Skip to content

GitLab

"lightx2v/models/vfi/vscode:/vscode.git/clone" did not exist on "c0b3601033a77329883bd87b5bdad481db517aed"
Commit aa58d024 authored by unknown's avatar unknown
Browse files

Initial add code.

parents
Hide whitespace changes
Inline Side-by-side
Showing with 2007 additions and 0 deletions
configs/recognition/i3d/README.md 0 → 100644
View file @ aa58d024
# I3D
[Quo Vadis, Action Recognition? A New Model and the Kinetics Dataset](https://openaccess.thecvf.com/content_cvpr_2017/html/Carreira_Quo_Vadis_Action_CVPR_2017_paper.html)
[Non-local Neural Networks](https://openaccess.thecvf.com/content_cvpr_2018/html/Wang_Non-Local_Neural_Networks_CVPR_2018_paper.html)
<!-- [ALGORITHM] -->
## Abstract
<!-- [ABSTRACT] -->
The paucity of videos in current action classification datasets (UCF-101 and HMDB-51) has made it difficult to identify good video architectures, as most methods obtain similar performance on existing small-scale benchmarks. This paper re-evaluates state-of-the-art architectures in light of the new Kinetics Human Action Video dataset. Kinetics has two orders of magnitude more data, with 400 human action classes and over 400 clips per class, and is collected from realistic, challenging YouTube videos. We provide an analysis on how current architectures fare on the task of action classification on this dataset and how much performance improves on the smaller benchmark datasets after pre-training on Kinetics. We also introduce a new Two-Stream Inflated 3D ConvNet (I3D) that is based on 2D ConvNet inflation: filters and pooling kernels of very deep image classification ConvNets are expanded into 3D, making it possible to learn seamless spatio-temporal feature extractors from video while leveraging successful ImageNet architecture designs and even their parameters. We show that, after pre-training on Kinetics, I3D models considerably improve upon the state-of-the-art in action classification, reaching 80.9% on HMDB-51 and 98.0% on UCF-101.
<!-- [IMAGE] -->
<div align=center>
<img src="https://user-images.githubusercontent.com/34324155/143043624-1944704a-5d3e-4a3f-b258-1505c49f6092.png" width="800"/>
</div>
## Results and Models
### Kinetics-400
| config | resolution | gpus | backbone | pretrain | top1 acc | top5 acc | inference_time(video/s) | gpu_mem(M) | ckpt | log | json |
| :----------------------------------------------------------------------------------------------------------------------------------------------- | :-------------: | :--: | :------: | :------: | :------: | :------: | :---------------------: | :--------: | :-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: | :-----------------------------------------------------------------------------------------------------------------------------------------: | :-----------------------------------------------------------------------------------------------------------------------------------------------: |
| [i3d_r50_32x2x1_100e_kinetics400_rgb](/configs/recognition/i3d/i3d_r50_32x2x1_100e_kinetics400_rgb.py) | 340x256 | 8 | ResNet50 | ImageNet | 72.68 | 90.78 | 1.7 (320x3 frames) | 5170 | [ckpt](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_32x2x1_100e_kinetics400_rgb/i3d_r50_32x2x1_100e_kinetics400_rgb_20200614-c25ef9a4.pth) | [log](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_32x2x1_100e_kinetics400_rgb/20200614_060456.log) | [json](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_32x2x1_100e_kinetics400_rgb/20200614_060456.log.json) |
| [i3d_r50_32x2x1_100e_kinetics400_rgb](/configs/recognition/i3d/i3d_r50_32x2x1_100e_kinetics400_rgb.py) | short-side 256 | 8 | ResNet50 | ImageNet | 73.27 | 90.92 | x | 5170 | [ckpt](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_256p_32x2x1_100e_kinetics400_rgb/i3d_r50_256p_32x2x1_100e_kinetics400_rgb_20200801-7d9f44de.pth) | [log](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_256p_32x2x1_100e_kinetics400_rgb/20200725_031555.log) | [json](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_256p_32x2x1_100e_kinetics400_rgb/20200725_031555.log.json) |
| [i3d_r50_video_32x2x1_100e_kinetics400_rgb](/configs/recognition/i3d/i3d_r50_video_32x2x1_100e_kinetics400_rgb.py) | short-side 256p | 8 | ResNet50 | ImageNet | 72.85 | 90.75 | x | 5170 | [ckpt](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_video_32x2x1_100e_kinetics400_rgb/i3d_r50_video_32x2x1_100e_kinetics400_rgb_20200826-e31c6f52.pth) | [log](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_video_32x2x1_100e_kinetics400_rgb/20200706_143014.log) | [json](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_video_32x2x1_100e_kinetics400_rgb/20200706_143014.log.json) |
| [i3d_r50_dense_32x2x1_100e_kinetics400_rgb](/configs/recognition/i3d/i3d_r50_dense_32x2x1_100e_kinetics400_rgb.py) | 340x256 | 8x2 | ResNet50 | ImageNet | 72.77 | 90.57 | 1.7 (320x3 frames) | 5170 | [ckpt](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_dense_32x2x1_100e_kinetics400_rgb/i3d_r50_dense_32x2x1_100e_kinetics400_rgb_20200616-2bbb4361.pth) | [log](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_dense_32x2x1_100e_kinetics400_rgb/20200616_230011.log) | [json](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_dense_32x2x1_100e_kinetics400_rgb/20200616_230011.log.json) |
| [i3d_r50_dense_32x2x1_100e_kinetics400_rgb](/configs/recognition/i3d/i3d_r50_dense_32x2x1_100e_kinetics400_rgb.py) | short-side 256 | 8 | ResNet50 | ImageNet | 73.48 | 91.00 | x | 5170 | [ckpt](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_dense_256p_32x2x1_100e_kinetics400_rgb/i3d_r50_dense_256p_32x2x1_100e_kinetics400_rgb_20200725-24eb54cc.pth) | [log](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_dense_256p_32x2x1_100e_kinetics400_rgb/20200725_031604.log) | [json](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_dense_256p_32x2x1_100e_kinetics400_rgb/20200725_031604.log.json) |
| [i3d_r50_lazy_32x2x1_100e_kinetics400_rgb](/configs/recognition/i3d/i3d_r50_lazy_32x2x1_100e_kinetics400_rgb.py) | 340x256 | 8 | ResNet50 | ImageNet | 72.32 | 90.72 | 1.8 (320x3 frames) | 5170 | [ckpt](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_fast_32x2x1_100e_kinetics400_rgb/i3d_r50_fast_32x2x1_100e_kinetics400_rgb_20200612-000e4d2a.pth) | [log](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_fast_32x2x1_100e_kinetics400_rgb/20200612_233836.log) | [json](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_fast_32x2x1_100e_kinetics400_rgb/20200612_233836.log.json) |
| [i3d_r50_lazy_32x2x1_100e_kinetics400_rgb](/configs/recognition/i3d/i3d_r50_lazy_32x2x1_100e_kinetics400_rgb.py) | short-side 256 | 8 | ResNet50 | ImageNet | 73.24 | 90.99 | x | 5170 | [ckpt](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_fast_256p_32x2x1_100e_kinetics400_rgb/i3d_r50_fast_256p_32x2x1_100e_kinetics400_rgb_20200817-4e90d1d5.pth) | [log](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_fast_256p_32x2x1_100e_kinetics400_rgb/20200725_031457.log) | [json](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_fast_256p_32x2x1_100e_kinetics400_rgb/20200725_031457.log.json) |
| [i3d_nl_embedded_gaussian_r50_32x2x1_100e_kinetics400_rgb](/configs/recognition/i3d/i3d_nl_embedded_gaussian_r50_32x2x1_100e_kinetics400_rgb.py) | short-side 256p | 8x4 | ResNet50 | ImageNet | 74.71 | 91.81 | x | 6438 | [ckpt](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_nl_embedded_gaussian_r50_32x2x1_100e_kinetics400_rgb/i3d_nl_embedded_gaussian_r50_32x2x1_100e_kinetics400_rgb_20200813-6e6aef1b.pth) | [log](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_nl_embedded_gaussian_r50_32x2x1_100e_kinetics400_rgb/20200813_034054.log) | [json](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_nl_embedded_gaussian_r50_32x2x1_100e_kinetics400_rgb/20200813_034054.log.json) |
| [i3d_nl_gaussian_r50_32x2x1_100e_kinetics400_rgb](/configs/recognition/i3d/i3d_nl_gaussian_r50_32x2x1_100e_kinetics400_rgb.py) | short-side 256p | 8x4 | ResNet50 | ImageNet | 73.37 | 91.26 | x | 4944 | [ckpt](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_nl_gaussian_r50_32x2x1_100e_kinetics400_rgb/i3d_nl_gaussian_r50_32x2x1_100e_kinetics400_rgb_20200815-17f84aa2.pth) | [log](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_nl_gaussian_r50_32x2x1_100e_kinetics400_rgb/20200813_034909.log) | [json](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_nl_gaussian_r50_32x2x1_100e_kinetics400_rgb/20200813_034909.log.json) |
| [i3d_nl_dot_product_r50_32x2x1_100e_kinetics400_rgb](/configs/recognition/i3d/i3d_nl_dot_product_r50_32x2x1_100e_kinetics400_rgb.py) | short-side 256p | 8x4 | ResNet50 | ImageNet | 73.92 | 91.59 | x | 4832 | [ckpt](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_nl_dot_product_r50_32x2x1_100e_kinetics400_rgb/i3d_nl_dot_product_r50_32x2x1_100e_kinetics400_rgb_20200814-7c30d5bb.pth) | [log](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_nl_dot_product_r50_32x2x1_100e_kinetics400_rgb/20200814_044208.log) | [json](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_nl_dot_product_r50_32x2x1_100e_kinetics400_rgb/20200814_044208.log.json) |
:::{note}
1. The **gpus** indicates the number of gpu we used to get the checkpoint. It is noteworthy that the configs we provide are used for 8 gpus as default.
According to the [Linear Scaling Rule](https://arxiv.org/abs/1706.02677), you may set the learning rate proportional to the batch size if you use different GPUs or videos per GPU,
e.g., lr=0.01 for 4 GPUs x 2 video/gpu and lr=0.08 for 16 GPUs x 4 video/gpu.
2. The **inference_time** is got by this [benchmark script](/tools/analysis/benchmark.py), where we use the sampling frames strategy of the test setting and only care about the model inference time, not including the IO time and pre-processing time. For each setting, we use 1 gpu and set batch size (videos per gpu) to 1 to calculate the inference time.
3. The validation set of Kinetics400 we used consists of 19796 videos. These videos are available at [Kinetics400-Validation](https://mycuhk-my.sharepoint.com/:u:/g/personal/1155136485_link_cuhk_edu_hk/EbXw2WX94J1Hunyt3MWNDJUBz-nHvQYhO9pvKqm6g39PMA?e=a9QldB). The corresponding [data list](https://download.openmmlab.com/mmaction/dataset/k400_val/kinetics_val_list.txt) (each line is of the format 'video_id, num_frames, label_index') and the [label map](https://download.openmmlab.com/mmaction/dataset/k400_val/kinetics_class2ind.txt) are also available.
:::
For more details on data preparation, you can refer to Kinetics400 in [Data Preparation](/docs/data_preparation.md).
## Train
You can use the following command to train a model.
```shell
python tools/train.py ${CONFIG_FILE} [optional arguments]
```
Example: train I3D model on Kinetics-400 dataset in a deterministic option with periodic validation.
```shell
python tools/train.py configs/recognition/i3d/i3d_r50_32x2x1_100e_kinetics400_rgb.py \
--work-dir work_dirs/i3d_r50_32x2x1_100e_kinetics400_rgb \
--validate --seed 0 --deterministic
```
For more details, you can refer to **Training setting** part in [getting_started](/docs/getting_started.md#training-setting).
## Test
You can use the following command to test a model.
```shell
python tools/test.py ${CONFIG_FILE} ${CHECKPOINT_FILE} [optional arguments]
```
Example: test I3D model on Kinetics-400 dataset and dump the result to a json file.
```shell
python tools/test.py configs/recognition/i3d/i3d_r50_32x2x1_100e_kinetics400_rgb.py \
checkpoints/SOME_CHECKPOINT.pth --eval top_k_accuracy mean_class_accuracy \
--out result.json --average-clips prob
```
For more details, you can refer to **Test a dataset** part in [getting_started](/docs/getting_started.md#test-a-dataset).
## Citation
```BibTeX
@inproceedings{inproceedings,
author = {Carreira, J. and Zisserman, Andrew},
year = {2017},
month = {07},
pages = {4724-4733},
title = {Quo Vadis, Action Recognition? A New Model and the Kinetics Dataset},
doi = {10.1109/CVPR.2017.502}
}
```
<!-- [BACKBONE] -->
```BibTeX
@article{NonLocal2018,
author = {Xiaolong Wang and Ross Girshick and Abhinav Gupta and Kaiming He},
title = {Non-local Neural Networks},
journal = {CVPR},
year = {2018}
}
```
configs/recognition/i3d/README_zh-CN.md 0 → 100644
View file @ aa58d024
# I3D
## 简介
<!-- [ALGORITHM] -->
```BibTeX
@inproceedings{inproceedings,
author = {Carreira, J. and Zisserman, Andrew},
year = {2017},
month = {07},
pages = {4724-4733},
title = {Quo Vadis, Action Recognition? A New Model and the Kinetics Dataset},
doi = {10.1109/CVPR.2017.502}
}
```
<!-- [BACKBONE] -->
```BibTeX
@article{NonLocal2018,
author = {Xiaolong Wang and Ross Girshick and Abhinav Gupta and Kaiming He},
title = {Non-local Neural Networks},
journal = {CVPR},
year = {2018}
}
```
## 模型库
### Kinetics-400
| 配置文件 | 分辨率 | GPU 数量 | 主干网络 | 预训练 | top1 准确率 | top5 准确率 | 推理时间 (video/s) | GPU 显存占用 (M) | ckpt | log | json |
| :----------------------------------------------------------------------------------------------------------------------------------------------- | :-------: | :------: | :------: | :------: | :---------: | :---------: | :----------------: | :--------------: | :-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: | :-----------------------------------------------------------------------------------------------------------------------------------------: | :-----------------------------------------------------------------------------------------------------------------------------------------------: |
| [i3d_r50_32x2x1_100e_kinetics400_rgb](/configs/recognition/i3d/i3d_r50_32x2x1_100e_kinetics400_rgb.py) | 340x256 | 8 | ResNet50 | ImageNet | 72.68 | 90.78 | 1.7 (320x3 frames) | 5170 | [ckpt](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_32x2x1_100e_kinetics400_rgb/i3d_r50_32x2x1_100e_kinetics400_rgb_20200614-c25ef9a4.pth) | [log](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_32x2x1_100e_kinetics400_rgb/20200614_060456.log) | [json](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_32x2x1_100e_kinetics400_rgb/20200614_060456.log.json) |
| [i3d_r50_32x2x1_100e_kinetics400_rgb](/configs/recognition/i3d/i3d_r50_32x2x1_100e_kinetics400_rgb.py) | 短边 256 | 8 | ResNet50 | ImageNet | 73.27 | 90.92 | x | 5170 | [ckpt](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_256p_32x2x1_100e_kinetics400_rgb/i3d_r50_256p_32x2x1_100e_kinetics400_rgb_20200801-7d9f44de.pth) | [log](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_256p_32x2x1_100e_kinetics400_rgb/20200725_031555.log) | [json](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_256p_32x2x1_100e_kinetics400_rgb/20200725_031555.log.json) |
| [i3d_r50_video_32x2x1_100e_kinetics400_rgb](/configs/recognition/i3d/i3d_r50_video_32x2x1_100e_kinetics400_rgb.py) | 短边 256p | 8 | ResNet50 | ImageNet | 72.85 | 90.75 | x | 5170 | [ckpt](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_video_32x2x1_100e_kinetics400_rgb/i3d_r50_video_32x2x1_100e_kinetics400_rgb_20200826-e31c6f52.pth) | [log](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_video_32x2x1_100e_kinetics400_rgb/20200706_143014.log) | [json](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_video_32x2x1_100e_kinetics400_rgb/20200706_143014.log.json) |
| [i3d_r50_dense_32x2x1_100e_kinetics400_rgb](/configs/recognition/i3d/i3d_r50_dense_32x2x1_100e_kinetics400_rgb.py) | 340x256 | 8x2 | ResNet50 | ImageNet | 72.77 | 90.57 | 1.7 (320x3 frames) | 5170 | [ckpt](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_dense_32x2x1_100e_kinetics400_rgb/i3d_r50_dense_32x2x1_100e_kinetics400_rgb_20200616-2bbb4361.pth) | [log](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_dense_32x2x1_100e_kinetics400_rgb/20200616_230011.log) | [json](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_dense_32x2x1_100e_kinetics400_rgb/20200616_230011.log.json) |
| [i3d_r50_dense_32x2x1_100e_kinetics400_rgb](/configs/recognition/i3d/i3d_r50_dense_32x2x1_100e_kinetics400_rgb.py) | 短边 256 | 8 | ResNet50 | ImageNet | 73.48 | 91.00 | x | 5170 | [ckpt](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_dense_256p_32x2x1_100e_kinetics400_rgb/i3d_r50_dense_256p_32x2x1_100e_kinetics400_rgb_20200725-24eb54cc.pth) | [log](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_dense_256p_32x2x1_100e_kinetics400_rgb/20200725_031604.log) | [json](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_dense_256p_32x2x1_100e_kinetics400_rgb/20200725_031604.log.json) |
| [i3d_r50_lazy_32x2x1_100e_kinetics400_rgb](/configs/recognition/i3d/i3d_r50_lazy_32x2x1_100e_kinetics400_rgb.py) | 340x256 | 8 | ResNet50 | ImageNet | 72.32 | 90.72 | 1.8 (320x3 frames) | 5170 | [ckpt](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_fast_32x2x1_100e_kinetics400_rgb/i3d_r50_fast_32x2x1_100e_kinetics400_rgb_20200612-000e4d2a.pth) | [log](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_fast_32x2x1_100e_kinetics400_rgb/20200612_233836.log) | [json](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_fast_32x2x1_100e_kinetics400_rgb/20200612_233836.log.json) |
| [i3d_r50_lazy_32x2x1_100e_kinetics400_rgb](/configs/recognition/i3d/i3d_r50_lazy_32x2x1_100e_kinetics400_rgb.py) | 短边 256 | 8 | ResNet50 | ImageNet | 73.24 | 90.99 | x | 5170 | [ckpt](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_fast_256p_32x2x1_100e_kinetics400_rgb/i3d_r50_fast_256p_32x2x1_100e_kinetics400_rgb_20200817-4e90d1d5.pth) | [log](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_fast_256p_32x2x1_100e_kinetics400_rgb/20200725_031457.log) | [json](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_fast_256p_32x2x1_100e_kinetics400_rgb/20200725_031457.log.json) |
| [i3d_nl_embedded_gaussian_r50_32x2x1_100e_kinetics400_rgb](/configs/recognition/i3d/i3d_nl_embedded_gaussian_r50_32x2x1_100e_kinetics400_rgb.py) | 短边 256p | 8x4 | ResNet50 | ImageNet | 74.71 | 91.81 | x | 6438 | [ckpt](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_nl_embedded_gaussian_r50_32x2x1_100e_kinetics400_rgb/i3d_nl_embedded_gaussian_r50_32x2x1_100e_kinetics400_rgb_20200813-6e6aef1b.pth) | [log](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_nl_embedded_gaussian_r50_32x2x1_100e_kinetics400_rgb/20200813_034054.log) | [json](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_nl_embedded_gaussian_r50_32x2x1_100e_kinetics400_rgb/20200813_034054.log.json) |
| [i3d_nl_gaussian_r50_32x2x1_100e_kinetics400_rgb](/configs/recognition/i3d/i3d_nl_gaussian_r50_32x2x1_100e_kinetics400_rgb.py) | 短边 256p | 8x4 | ResNet50 | ImageNet | 73.37 | 91.26 | x | 4944 | [ckpt](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_nl_gaussian_r50_32x2x1_100e_kinetics400_rgb/i3d_nl_gaussian_r50_32x2x1_100e_kinetics400_rgb_20200815-17f84aa2.pth) | [log](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_nl_gaussian_r50_32x2x1_100e_kinetics400_rgb/20200813_034909.log) | [json](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_nl_gaussian_r50_32x2x1_100e_kinetics400_rgb/20200813_034909.log.json) |
| [i3d_nl_dot_product_r50_32x2x1_100e_kinetics400_rgb](/configs/recognition/i3d/i3d_nl_dot_product_r50_32x2x1_100e_kinetics400_rgb.py) | 短边 256p | 8x4 | ResNet50 | ImageNet | 73.92 | 91.59 | x | 4832 | [ckpt](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_nl_dot_product_r50_32x2x1_100e_kinetics400_rgb/i3d_nl_dot_product_r50_32x2x1_100e_kinetics400_rgb_20200814-7c30d5bb.pth) | [log](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_nl_dot_product_r50_32x2x1_100e_kinetics400_rgb/20200814_044208.log) | [json](https://download.openmmlab.com/mmaction/recognition/i3d/i3d_nl_dot_product_r50_32x2x1_100e_kinetics400_rgb/20200814_044208.log.json) |
注：
1. 这里的 **GPU 数量** 指的是得到模型权重文件对应的 GPU 个数。默认地，MMAction2 所提供的配置文件对应使用 8 块 GPU 进行训练的情况。
依据 [线性缩放规则](https://arxiv.org/abs/1706.02677)，当用户使用不同数量的 GPU 或者每块 GPU 处理不同视频个数时，需要根据批大小等比例地调节学习率。
如，lr=0.01 对应 4 GPUs x 2 video/gpu，以及 lr=0.08 对应 16 GPUs x 4 video/gpu。
2. 这里的 **推理时间** 是根据 [基准测试脚本](/tools/analysis/benchmark.py) 获得的，采用测试时的采帧策略，且只考虑模型的推理时间，
并不包括 IO 时间以及预处理时间。对于每个配置，MMAction2 使用 1 块 GPU 并设置批大小（每块 GPU 处理的视频个数）为 1 来计算推理时间。
3. 我们使用的 Kinetics400 验证集包含 19796 个视频，用户可以从 [验证集视频](https://mycuhk-my.sharepoint.com/:u:/g/personal/1155136485_link_cuhk_edu_hk/EbXw2WX94J1Hunyt3MWNDJUBz-nHvQYhO9pvKqm6g39PMA?e=a9QldB) 下载这些视频。同时也提供了对应的 [数据列表](https://download.openmmlab.com/mmaction/dataset/k400_val/kinetics_val_list.txt) （每行格式为：视频 ID，视频帧数目，类别序号）以及 [标签映射](https://download.openmmlab.com/mmaction/dataset/k400_val/kinetics_class2ind.txt) （类别序号到类别名称）。
对于数据集准备的细节，用户可参考 [数据集准备文档](/docs_zh_CN/data_preparation.md) 中的 Kinetics400 部分。
## 如何训练
用户可以使用以下指令进行模型训练。
```shell
python tools/train.py ${CONFIG_FILE} [optional arguments]
```
例如：以一个确定性的训练方式，辅以定期的验证过程进行 I3D 模型在 Kinetics400 数据集上的训练。
```shell
python tools/train.py configs/recognition/i3d/i3d_r50_32x2x1_100e_kinetics400_rgb.py \
--work-dir work_dirs/i3d_r50_32x2x1_100e_kinetics400_rgb \
--validate --seed 0 --deterministic
```
更多训练细节，可参考 [基础教程](/docs_zh_CN/getting_started.md#%E8%AE%AD%E7%BB%83%E9%85%8D%E7%BD%AE) 中的 **训练配置** 部分。
## 如何测试
用户可以使用以下指令进行模型测试。
```shell
python tools/test.py ${CONFIG_FILE} ${CHECKPOINT_FILE} [optional arguments]
```
例如：在 Kinetics400 数据集上测试 I3D 模型，并将结果导出为一个 json 文件。
```shell
python tools/test.py configs/recognition/i3d/i3d_r50_32x2x1_100e_kinetics400_rgb.py \
checkpoints/SOME_CHECKPOINT.pth --eval top_k_accuracy mean_class_accuracy \
--out result.json --average-clips prob
```
更多测试细节，可参考 [基础教程](/docs_zh_CN/getting_started.md#%E6%B5%8B%E8%AF%95%E6%9F%90%E4%B8%AA%E6%95%B0%E6%8D%AE%E9%9B%86) 中的 **测试某个数据集** 部分。
configs/recognition/i3d/i3d_nl_dot_product_r50_32x2x1_100e_kinetics400_rgb.py 0 → 100644
View file @ aa58d024
_base_ = [
'../../_base_/models/i3d_r50.py', '../../_base_/schedules/sgd_100e.py',
'../../_base_/default_runtime.py'
]
# model settings
model = dict(
backbone=dict(
non_local=((0, 0, 0), (0, 1, 0, 1), (0, 1, 0, 1, 0, 1), (0, 0, 0)),
non_local_cfg=dict(
sub_sample=True,
use_scale=False,
norm_cfg=dict(type='BN3d', requires_grad=True),
mode='dot_product')))
# dataset settings
dataset_type = 'RawframeDataset'
data_root = 'data/kinetics400/rawframes_train'
data_root_val = 'data/kinetics400/rawframes_val'
ann_file_train = 'data/kinetics400/kinetics400_train_list_rawframes.txt'
ann_file_val = 'data/kinetics400/kinetics400_val_list_rawframes.txt'
ann_file_test = 'data/kinetics400/kinetics400_val_list_rawframes.txt'
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_bgr=False)
train_pipeline = [
dict(type='SampleFrames', clip_len=32, frame_interval=2, num_clips=1),
dict(type='RawFrameDecode'),
dict(type='Resize', scale=(-1, 256)),
dict(
type='MultiScaleCrop',
input_size=224,
scales=(1, 0.8),
random_crop=False,
max_wh_scale_gap=0),
dict(type='Resize', scale=(224, 224), keep_ratio=False),
dict(type='Flip', flip_ratio=0.5),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs', 'label'])
]
val_pipeline = [
dict(
type='SampleFrames',
clip_len=32,
frame_interval=2,
num_clips=1,
test_mode=True),
dict(type='RawFrameDecode'),
dict(type='Resize', scale=(-1, 256)),
dict(type='CenterCrop', crop_size=224),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs'])
]
test_pipeline = [
dict(
type='SampleFrames',
clip_len=32,
frame_interval=2,
num_clips=10,
test_mode=True),
dict(type='RawFrameDecode'),
dict(type='Resize', scale=(-1, 256)),
dict(type='ThreeCrop', crop_size=256),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs'])
]
data = dict(
videos_per_gpu=8,
workers_per_gpu=2,
test_dataloader=dict(videos_per_gpu=1),
train=dict(
type=dataset_type,
ann_file=ann_file_train,
data_prefix=data_root,
pipeline=train_pipeline),
val=dict(
type=dataset_type,
ann_file=ann_file_val,
data_prefix=data_root_val,
pipeline=val_pipeline),
test=dict(
type=dataset_type,
ann_file=ann_file_val,
data_prefix=data_root_val,
pipeline=test_pipeline))
evaluation = dict(
interval=5, metrics=['top_k_accuracy', 'mean_class_accuracy'])
# runtime settings
checkpoint_config = dict(interval=5)
work_dir = './work_dirs/i3d_nl_dot_product_r50_32x2x1_100e_kinetics400_rgb/'
configs/recognition/i3d/i3d_nl_embedded_gaussian_r50_32x2x1_100e_kinetics400_rgb.py 0 → 100644
View file @ aa58d024
_base_ = ['./i3d_nl_dot_product_r50_32x2x1_100e_kinetics400_rgb.py']
# model settings
model = dict(
backbone=dict(
non_local_cfg=dict(
sub_sample=True,
use_scale=False,
norm_cfg=dict(type='BN3d', requires_grad=True),
mode='embedded_gaussian')))
# runtime settings
work_dir = './work_dirs/i3d_nl_embedded_gaussian_r50_32x2x1_100e_kinetics400_rgb/' # noqa: E501
configs/recognition/i3d/i3d_nl_gaussian_r50_32x2x1_100e_kinetics400_rgb.py 0 → 100644
View file @ aa58d024
_base_ = ['./i3d_nl_dot_product_r50_32x2x1_100e_kinetics400_rgb.py']
# model settings
model = dict(
backbone=dict(
non_local_cfg=dict(
sub_sample=True,
use_scale=False,
norm_cfg=dict(type='BN3d', requires_grad=True),
mode='gaussian')))
# runtime settings
work_dir = './work_dirs/i3d_nl_gaussian_r50_32x2x1_100e_kinetics400_rgb/'
configs/recognition/i3d/i3d_r50_32x2x1_100e_kinetics400_rgb.py 0 → 100644
View file @ aa58d024
_base_ = [
'../../_base_/models/i3d_r50.py', '../../_base_/schedules/sgd_100e.py',
'../../_base_/default_runtime.py'
]
# dataset settings
dataset_type = 'RawframeDataset'
data_root = 'data/kinetics400/rawframes_train'
data_root_val = 'data/kinetics400/rawframes_val'
ann_file_train = 'data/kinetics400/kinetics400_train_list_rawframes.txt'
ann_file_val = 'data/kinetics400/kinetics400_val_list_rawframes.txt'
ann_file_test = 'data/kinetics400/kinetics400_val_list_rawframes.txt'
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_bgr=False)
train_pipeline = [
dict(type='SampleFrames', clip_len=32, frame_interval=2, num_clips=1),
dict(type='RawFrameDecode'),
dict(type='Resize', scale=(-1, 256)),
dict(
type='MultiScaleCrop',
input_size=224,
scales=(1, 0.8),
random_crop=False,
max_wh_scale_gap=0),
dict(type='Resize', scale=(224, 224), keep_ratio=False),
dict(type='Flip', flip_ratio=0.5),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs', 'label'])
]
val_pipeline = [
dict(
type='SampleFrames',
clip_len=32,
frame_interval=2,
num_clips=1,
test_mode=True),
dict(type='RawFrameDecode'),
dict(type='Resize', scale=(-1, 256)),
dict(type='CenterCrop', crop_size=224),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs'])
]
test_pipeline = [
dict(
type='SampleFrames',
clip_len=32,
frame_interval=2,
num_clips=10,
test_mode=True),
dict(type='RawFrameDecode'),
dict(type='Resize', scale=(-1, 256)),
dict(type='ThreeCrop', crop_size=256),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs'])
]
data = dict(
videos_per_gpu=8,
workers_per_gpu=2,
test_dataloader=dict(videos_per_gpu=1),
train=dict(
type=dataset_type,
ann_file=ann_file_train,
data_prefix=data_root,
pipeline=train_pipeline),
val=dict(
type=dataset_type,
ann_file=ann_file_val,
data_prefix=data_root_val,
pipeline=val_pipeline),
test=dict(
type=dataset_type,
ann_file=ann_file_val,
data_prefix=data_root_val,
pipeline=test_pipeline))
evaluation = dict(
interval=5, metrics=['top_k_accuracy', 'mean_class_accuracy'])
# runtime settings
checkpoint_config = dict(interval=5)
work_dir = './work_dirs/i3d_r50_32x2x1_100e_kinetics400_rgb/'
configs/recognition/i3d/i3d_r50_dense_32x2x1_100e_kinetics400_rgb.py 0 → 100644
View file @ aa58d024
_base_ = ['./i3d_r50_32x2x1_100e_kinetics400_rgb.py']
# dataset settings
dataset_type = 'RawframeDataset'
data_root = 'data/kinetics400/rawframes_train'
data_root_val = 'data/kinetics400/rawframes_val'
ann_file_train = 'data/kinetics400/kinetics400_train_list_rawframes.txt'
ann_file_val = 'data/kinetics400/kinetics400_val_list_rawframes.txt'
ann_file_test = 'data/kinetics400/kinetics400_val_list_rawframes.txt'
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_bgr=False)
train_pipeline = [
dict(type='DenseSampleFrames', clip_len=32, frame_interval=2, num_clips=1),
dict(type='RawFrameDecode'),
dict(type='Resize', scale=(-1, 256)),
dict(
type='MultiScaleCrop',
input_size=224,
scales=(1, 0.8),
random_crop=False,
max_wh_scale_gap=0),
dict(type='Resize', scale=(224, 224), keep_ratio=False),
dict(type='Flip', flip_ratio=0.5),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs', 'label'])
]
val_pipeline = [
dict(
type='DenseSampleFrames',
clip_len=32,
frame_interval=2,
num_clips=1,
test_mode=True),
dict(type='RawFrameDecode'),
dict(type='Resize', scale=(-1, 256)),
dict(type='CenterCrop', crop_size=224),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs'])
]
test_pipeline = [
dict(
type='DenseSampleFrames',
clip_len=32,
frame_interval=2,
num_clips=1,
test_mode=True),
dict(type='RawFrameDecode'),
dict(type='Resize', scale=(-1, 256)),
dict(type='ThreeCrop', crop_size=256),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs'])
]
data = dict(
videos_per_gpu=8,
workers_per_gpu=2,
test_dataloader=dict(videos_per_gpu=1),
train=dict(
type=dataset_type,
ann_file=ann_file_train,
data_prefix=data_root,
pipeline=train_pipeline),
val=dict(
type=dataset_type,
ann_file=ann_file_val,
data_prefix=data_root_val,
pipeline=val_pipeline),
test=dict(
type=dataset_type,
ann_file=ann_file_val,
data_prefix=data_root_val,
pipeline=test_pipeline))
# runtime settings
work_dir = './work_dirs/i3d_r50_dense_32x2x1_100e_kinetics400_rgb/'
configs/recognition/i3d/i3d_r50_heavy_8x8x1_100e_kinetics400_rgb.py 0 → 100644
View file @ aa58d024
_base_ = ['./i3d_r50_32x2x1_100e_kinetics400_rgb.py']
# model settings
model = dict(
backbone=dict(
inflate=(1, 1, 1, 1),
conv1_stride_t=1,
pool1_stride_t=1,
with_pool2=True))
# dataset settings
dataset_type = 'RawframeDataset'
data_root = 'data/kinetics400/rawframes_train'
data_root_val = 'data/kinetics400/rawframes_val'
ann_file_train = 'data/kinetics400/kinetics400_train_list_rawframes.txt'
ann_file_val = 'data/kinetics400/kinetics400_val_list_rawframes.txt'
ann_file_test = 'data/kinetics400/kinetics400_val_list_rawframes.txt'
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_bgr=False)
train_pipeline = [
dict(type='SampleFrames', clip_len=8, frame_interval=8, num_clips=1),
dict(type='RawFrameDecode'),
dict(type='Resize', scale=(-1, 256)),
dict(
type='MultiScaleCrop',
input_size=224,
scales=(1, 0.8),
random_crop=False,
max_wh_scale_gap=0),
dict(type='Resize', scale=(224, 224), keep_ratio=False),
dict(type='Flip', flip_ratio=0.5),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs', 'label'])
]
val_pipeline = [
dict(
type='SampleFrames',
clip_len=8,
frame_interval=8,
num_clips=1,
test_mode=True),
dict(type='RawFrameDecode'),
dict(type='Resize', scale=(-1, 256)),
dict(type='CenterCrop', crop_size=224),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs'])
]
test_pipeline = [
dict(
type='SampleFrames',
clip_len=8,
frame_interval=8,
num_clips=10,
test_mode=True),
dict(type='RawFrameDecode'),
dict(type='Resize', scale=(-1, 256)),
dict(type='ThreeCrop', crop_size=256),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs'])
]
data = dict(
videos_per_gpu=8,
workers_per_gpu=2,
test_dataloader=dict(videos_per_gpu=1),
train=dict(
type=dataset_type,
ann_file=ann_file_train,
data_prefix=data_root,
pipeline=train_pipeline),
val=dict(
type=dataset_type,
ann_file=ann_file_val,
data_prefix=data_root_val,
pipeline=val_pipeline),
test=dict(
type=dataset_type,
ann_file=ann_file_val,
data_prefix=data_root_val,
pipeline=test_pipeline))
# runtime settings
work_dir = './work_dirs/i3d_r50_heavy_8x8x1_100e_kinetics400_rgb/'
configs/recognition/i3d/i3d_r50_lazy_32x2x1_100e_kinetics400_rgb.py 0 → 100644
View file @ aa58d024
_base_ = ['./i3d_r50_32x2x1_100e_kinetics400_rgb.py']
# dataset settings
dataset_type = 'RawframeDataset'
data_root = 'data/kinetics400/rawframes_train'
data_root_val = 'data/kinetics400/rawframes_val'
ann_file_train = 'data/kinetics400/kinetics400_train_list_rawframes.txt'
ann_file_val = 'data/kinetics400/kinetics400_val_list_rawframes.txt'
ann_file_test = 'data/kinetics400/kinetics400_val_list_rawframes.txt'
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_bgr=False)
train_pipeline = [
dict(type='SampleFrames', clip_len=32, frame_interval=2, num_clips=1),
dict(type='RawFrameDecode', decoding_backend='turbojpeg'),
dict(type='Resize', scale=(-1, 256), lazy=True),
dict(
type='MultiScaleCrop',
input_size=224,
scales=(1, 0.8),
random_crop=False,
max_wh_scale_gap=0,
lazy=True),
dict(type='Resize', scale=(224, 224), keep_ratio=False, lazy=True),
dict(type='Flip', flip_ratio=0.5, lazy=True),
dict(type='Fuse'),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs', 'label'])
]
val_pipeline = [
dict(
type='SampleFrames',
clip_len=32,
frame_interval=2,
num_clips=1,
test_mode=True),
dict(type='RawFrameDecode', decoding_backend='turbojpeg'),
dict(type='Resize', scale=(-1, 256), lazy=True),
dict(type='CenterCrop', crop_size=224, lazy=True),
dict(type='Flip', flip_ratio=0, lazy=True),
dict(type='Fuse'),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs'])
]
test_pipeline = [
dict(
type='SampleFrames',
clip_len=32,
frame_interval=2,
num_clips=10,
test_mode=True),
dict(type='RawFrameDecode', decoding_backend='turbojpeg'),
dict(type='Resize', scale=(-1, 256)),
dict(type='ThreeCrop', crop_size=256),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs'])
]
data = dict(
videos_per_gpu=8,
workers_per_gpu=2,
test_dataloader=dict(videos_per_gpu=1),
train=dict(
type=dataset_type,
ann_file=ann_file_train,
data_prefix=data_root,
pipeline=train_pipeline),
val=dict(
type=dataset_type,
ann_file=ann_file_val,
data_prefix=data_root_val,
pipeline=val_pipeline),
test=dict(
type=dataset_type,
ann_file=ann_file_val,
data_prefix=data_root_val,
pipeline=test_pipeline))
# runtime settings
work_dir = './work_dirs/i3d_r50_lazy_32x2x1_100e_kinetics400_rgb/'
configs/recognition/i3d/i3d_r50_video_32x2x1_100e_kinetics400_rgb.py 0 → 100644
View file @ aa58d024
_base_ = ['./i3d_r50_32x2x1_100e_kinetics400_rgb.py']
# dataset settings
dataset_type = 'VideoDataset'
data_root = 'data/kinetics400/videos_train'
data_root_val = 'data/kinetics400/videos_val'
ann_file_train = 'data/kinetics400/kinetics400_train_list_videos.txt'
ann_file_val = 'data/kinetics400/kinetics400_val_list_videos.txt'
ann_file_test = 'data/kinetics400/kinetics400_val_list_videos.txt'
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_bgr=False)
train_pipeline = [
dict(type='DecordInit'),
dict(type='SampleFrames', clip_len=32, frame_interval=2, num_clips=1),
dict(type='DecordDecode'),
dict(type='Resize', scale=(-1, 256)),
dict(
type='MultiScaleCrop',
input_size=224,
scales=(1, 0.8),
random_crop=False,
max_wh_scale_gap=0),
dict(type='Resize', scale=(224, 224), keep_ratio=False),
dict(type='Flip', flip_ratio=0.5),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs', 'label'])
]
val_pipeline = [
dict(type='DecordInit'),
dict(
type='SampleFrames',
clip_len=32,
frame_interval=2,
num_clips=1,
test_mode=True),
dict(type='DecordDecode'),
dict(type='Resize', scale=(-1, 256)),
dict(type='CenterCrop', crop_size=224),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs'])
]
test_pipeline = [
dict(type='DecordInit'),
dict(
type='SampleFrames',
clip_len=32,
frame_interval=2,
num_clips=10,
test_mode=True),
dict(type='DecordDecode'),
dict(type='Resize', scale=(-1, 256)),
dict(type='ThreeCrop', crop_size=256),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs'])
]
data = dict(
videos_per_gpu=8,
workers_per_gpu=2,
test_dataloader=dict(videos_per_gpu=1),
train=dict(
type=dataset_type,
ann_file=ann_file_train,
data_prefix=data_root,
pipeline=train_pipeline),
val=dict(
type=dataset_type,
ann_file=ann_file_val,
data_prefix=data_root_val,
pipeline=val_pipeline),
test=dict(
type=dataset_type,
ann_file=ann_file_val,
data_prefix=data_root_val,
pipeline=test_pipeline))
# runtime settings
work_dir = './work_dirs/i3d_r50_video_3d_32x2x1_100e_kinetics400_rgb/'
configs/recognition/i3d/i3d_r50_video_heavy_8x8x1_100e_kinetics400_rgb.py 0 → 100644
View file @ aa58d024
_base_ = ['./i3d_r50_heavy_8x8x1_100e_kinetics400_rgb.py']
# dataset settings
dataset_type = 'VideoDataset'
data_root = 'data/kinetics400/videos_train'
data_root_val = 'data/kinetics400/videos_val'
ann_file_train = 'data/kinetics400/kinetics400_train_list_videos.txt'
ann_file_val = 'data/kinetics400/kinetics400_val_list_videos.txt'
ann_file_test = 'data/kinetics400/kinetics400_val_list_videos.txt'
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_bgr=False)
train_pipeline = [
dict(type='DecordInit'),
dict(type='SampleFrames', clip_len=8, frame_interval=8, num_clips=1),
dict(type='DecordDecode'),
dict(type='Resize', scale=(-1, 256)),
dict(
type='MultiScaleCrop',
input_size=224,
scales=(1, 0.8),
random_crop=False,
max_wh_scale_gap=0),
dict(type='Resize', scale=(224, 224), keep_ratio=False),
dict(type='Flip', flip_ratio=0.5),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs', 'label'])
]
val_pipeline = [
dict(type='DecordInit'),
dict(
type='SampleFrames',
clip_len=8,
frame_interval=8,
num_clips=1,
test_mode=True),
dict(type='DecordDecode'),
dict(type='Resize', scale=(-1, 256)),
dict(type='CenterCrop', crop_size=224),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs'])
]
test_pipeline = [
dict(type='DecordInit'),
dict(
type='SampleFrames',
clip_len=8,
frame_interval=8,
num_clips=10,
test_mode=True),
dict(type='DecordDecode'),
dict(type='Resize', scale=(-1, 256)),
dict(type='ThreeCrop', crop_size=256),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs'])
]
data = dict(
videos_per_gpu=8,
workers_per_gpu=2,
test_dataloader=dict(videos_per_gpu=1),
train=dict(
type=dataset_type,
ann_file=ann_file_train,
data_prefix=data_root,
pipeline=train_pipeline),
val=dict(
type=dataset_type,
ann_file=ann_file_val,
data_prefix=data_root_val,
pipeline=val_pipeline),
test=dict(
type=dataset_type,
ann_file=ann_file_val,
data_prefix=data_root_val,
pipeline=test_pipeline))
#
# runtime settings
work_dir = './work_dirs/i3d_r50_video_heavy_8x8x1_100e_kinetics400_rgb/'
configs/recognition/i3d/i3d_r50_video_imgaug_32x2x1_100e_kinetics400_rgb.py 0 → 100644
View file @ aa58d024
_base_ = ['../../_base_/models/i3d_r50.py']
# dataset settings
dataset_type = 'VideoDataset'
data_root = 'data/kinetics400/videos_train'
data_root_val = 'data/kinetics400/videos_val'
ann_file_train = 'data/kinetics400/kinetics400_train_list_videos.txt'
ann_file_val = 'data/kinetics400/kinetics400_val_list_videos.txt'
ann_file_test = 'data/kinetics400/kinetics400_val_list_videos.txt'
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_bgr=False)
train_pipeline = [
dict(type='DecordInit'),
dict(type='SampleFrames', clip_len=32, frame_interval=2, num_clips=1),
dict(type='DecordDecode'),
dict(type='Resize', scale=(-1, 256)),
dict(
type='MultiScaleCrop',
input_size=224,
scales=(1, 0.8),
random_crop=False,
max_wh_scale_gap=0),
dict(type='Resize', scale=(224, 224), keep_ratio=False),
dict(
type='Imgaug',
transforms=[
dict(type='Fliplr', p=0.5),
dict(type='Rotate', rotate=(-20, 20)),
dict(type='Dropout', p=(0, 0.05))
]),
# dict(type='Imgaug', transforms='default'),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs', 'label'])
]
val_pipeline = [
dict(type='DecordInit'),
dict(
type='SampleFrames',
clip_len=32,
frame_interval=2,
num_clips=1,
test_mode=True),
dict(type='DecordDecode'),
dict(type='Resize', scale=(-1, 256)),
dict(type='CenterCrop', crop_size=224),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs'])
]
test_pipeline = [
dict(type='DecordInit'),
dict(
type='SampleFrames',
clip_len=32,
frame_interval=2,
num_clips=10,
test_mode=True),
dict(type='DecordDecode'),
dict(type='Resize', scale=(-1, 256)),
dict(type='ThreeCrop', crop_size=256),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs'])
]
data = dict(
videos_per_gpu=8,
workers_per_gpu=2,
test_dataloader=dict(videos_per_gpu=1),
train=dict(
type=dataset_type,
ann_file=ann_file_train,
data_prefix=data_root,
pipeline=train_pipeline),
val=dict(
type=dataset_type,
ann_file=ann_file_val,
data_prefix=data_root_val,
pipeline=val_pipeline),
test=dict(
type=dataset_type,
ann_file=ann_file_val,
data_prefix=data_root_val,
pipeline=test_pipeline))
# optimizer
optimizer = dict(
type='SGD', lr=0.01, momentum=0.9,
weight_decay=0.0001) # this lr is used for 8 gpus
optimizer_config = dict(grad_clip=dict(max_norm=40, norm_type=2))
# learning policy
lr_config = dict(policy='step', step=[40, 80])
total_epochs = 100
checkpoint_config = dict(interval=5)
evaluation = dict(
interval=5, metrics=['top_k_accuracy', 'mean_class_accuracy'])
log_config = dict(
interval=20,
hooks=[
dict(type='TextLoggerHook'),
# dict(type='TensorboardLoggerHook'),
])
# runtime settings
dist_params = dict(backend='nccl')
log_level = 'INFO'
work_dir = './work_dirs/i3d_r50_video_3d_32x2x1_100e_kinetics400_rgb/'
load_from = None
resume_from = None
workflow = [('train', 1)]
configs/recognition/i3d/i3d_r50_video_inference_32x2x1_100e_kinetics400_rgb.py 0 → 100644
View file @ aa58d024
_base_ = ['../../_base_/models/i3d_r50.py']
# dataset settings
dataset_type = 'VideoDataset'
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_bgr=False)
test_pipeline = [
dict(type='DecordInit', num_threads=1),
dict(
type='SampleFrames',
clip_len=32,
frame_interval=2,
num_clips=1,
test_mode=True),
dict(type='DecordDecode'),
dict(type='Resize', scale=(-1, 256)),
dict(type='ThreeCrop', crop_size=256),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs'])
]
data = dict(
videos_per_gpu=1,
workers_per_gpu=2,
test=dict(
type=dataset_type,
ann_file=None,
data_prefix=None,
pipeline=test_pipeline))
configs/recognition/i3d/metafile.yml 0 → 100644
View file @ aa58d024
Collections:
- Name: I3D
README: configs/recognition/i3d/README.md
Paper:
URL: https://arxiv.org/abs/1705.07750
Title: Quo Vadis, Action Recognition? A New Model and the Kinetics Dataset
Models:
- Config: configs/recognition/i3d/i3d_r50_32x2x1_100e_kinetics400_rgb.py
In Collection: I3D
Metadata:
Architecture: ResNet50
Batch Size: 8
Epochs: 100
FLOPs: 43564040192
Parameters: 28043472
Pretrained: ImageNet
Resolution: 340x256
Training Data: Kinetics-400
Training Resources: 8 GPUs
Modality: RGB
Name: i3d_r50_32x2x1_100e_kinetics400_rgb
Results:
- Dataset: Kinetics-400
Metrics:
Top 1 Accuracy: 72.68
Top 5 Accuracy: 90.78
Task: Action Recognition
Training Json Log: https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_32x2x1_100e_kinetics400_rgb/20200614_060456.log.json
Training Log: https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_32x2x1_100e_kinetics400_rgb/20200614_060456.log
Weights: https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_32x2x1_100e_kinetics400_rgb/i3d_r50_32x2x1_100e_kinetics400_rgb_20200614-c25ef9a4.pth
- Config: configs/recognition/i3d/i3d_r50_32x2x1_100e_kinetics400_rgb.py
In Collection: I3D
Metadata:
Architecture: ResNet50
Batch Size: 8
Epochs: 100
FLOPs: 43564040192
Parameters: 28043472
Pretrained: ImageNet
Resolution: short-side 256
Training Data: Kinetics-400
Training Resources: 8 GPUs
Modality: RGB
Name: i3d_r50_32x2x1_100e_kinetics400_rgb
Results:
- Dataset: Kinetics-400
Metrics:
Top 1 Accuracy: 73.27
Top 5 Accuracy: 90.92
Task: Action Recognition
Training Json Log: https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_256p_32x2x1_100e_kinetics400_rgb/20200725_031555.log.json
Training Log: https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_256p_32x2x1_100e_kinetics400_rgb/20200725_031555.log
Weights: https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_256p_32x2x1_100e_kinetics400_rgb/i3d_r50_256p_32x2x1_100e_kinetics400_rgb_20200801-7d9f44de.pth
- Config: configs/recognition/i3d/i3d_r50_video_32x2x1_100e_kinetics400_rgb.py
In Collection: I3D
Metadata:
Architecture: ResNet50
Batch Size: 8
Epochs: 100
FLOPs: 43564040192
Parameters: 28043472
Pretrained: ImageNet
Resolution: short-side 256p
Training Data: Kinetics-400
Training Resources: 8 GPUs
Modality: RGB
Name: i3d_r50_video_32x2x1_100e_kinetics400_rgb
Results:
- Dataset: Kinetics-400
Metrics:
Top 1 Accuracy: 72.85
Top 5 Accuracy: 90.75
Task: Action Recognition
Training Json Log: https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_video_32x2x1_100e_kinetics400_rgb/20200706_143014.log.json
Training Log: https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_video_32x2x1_100e_kinetics400_rgb/20200706_143014.log
Weights: https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_video_32x2x1_100e_kinetics400_rgb/i3d_r50_video_32x2x1_100e_kinetics400_rgb_20200826-e31c6f52.pth
- Config: configs/recognition/i3d/i3d_r50_dense_32x2x1_100e_kinetics400_rgb.py
In Collection: I3D
Metadata:
Architecture: ResNet50
Batch Size: 8
Epochs: 100
FLOPs: 43564040192
Parameters: 28043472
Pretrained: ImageNet
Resolution: 340x256
Training Data: Kinetics-400
Training Resources: 16 GPUs
Modality: RGB
Name: i3d_r50_dense_32x2x1_100e_kinetics400_rgb
Results:
- Dataset: Kinetics-400
Metrics:
Top 1 Accuracy: 72.77
Top 5 Accuracy: 90.57
Task: Action Recognition
Training Json Log: https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_dense_32x2x1_100e_kinetics400_rgb/20200616_230011.log.json
Training Log: https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_dense_32x2x1_100e_kinetics400_rgb/20200616_230011.log
Weights: https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_dense_32x2x1_100e_kinetics400_rgb/i3d_r50_dense_32x2x1_100e_kinetics400_rgb_20200616-2bbb4361.pth
- Config: configs/recognition/i3d/i3d_r50_dense_32x2x1_100e_kinetics400_rgb.py
In Collection: I3D
Metadata:
Architecture: ResNet50
Batch Size: 8
Epochs: 100
FLOPs: 43564040192
Parameters: 28043472
Pretrained: ImageNet
Resolution: short-side 256
Training Data: Kinetics-400
Training Resources: 8 GPUs
Modality: RGB
Name: i3d_r50_dense_32x2x1_100e_kinetics400_rgb
Results:
- Dataset: Kinetics-400
Metrics:
Top 1 Accuracy: 73.48
Top 5 Accuracy: 91.0
Task: Action Recognition
Training Json Log: https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_dense_256p_32x2x1_100e_kinetics400_rgb/20200725_031604.log.json
Training Log: https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_dense_256p_32x2x1_100e_kinetics400_rgb/20200725_031604.log
Weights: https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_dense_256p_32x2x1_100e_kinetics400_rgb/i3d_r50_dense_256p_32x2x1_100e_kinetics400_rgb_20200725-24eb54cc.pth
- Config: configs/recognition/i3d/i3d_r50_lazy_32x2x1_100e_kinetics400_rgb.py
In Collection: I3D
Metadata:
Architecture: ResNet50
Batch Size: 8
Epochs: 100
FLOPs: 43564040192
Parameters: 28043472
Pretrained: ImageNet
Resolution: 340x256
Training Data: Kinetics-400
Training Resources: 8 GPUs
Modality: RGB
Name: i3d_r50_lazy_32x2x1_100e_kinetics400_rgb
Results:
- Dataset: Kinetics-400
Metrics:
Top 1 Accuracy: 72.32
Top 5 Accuracy: 90.72
Task: Action Recognition
Training Json Log: https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_fast_32x2x1_100e_kinetics400_rgb/20200612_233836.log.json
Training Log: https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_fast_32x2x1_100e_kinetics400_rgb/20200612_233836.log
Weights: https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_fast_32x2x1_100e_kinetics400_rgb/i3d_r50_fast_32x2x1_100e_kinetics400_rgb_20200612-000e4d2a.pth
- Config: configs/recognition/i3d/i3d_r50_lazy_32x2x1_100e_kinetics400_rgb.py
In Collection: I3D
Metadata:
Architecture: ResNet50
Batch Size: 8
Epochs: 100
FLOPs: 43564040192
Parameters: 28043472
Pretrained: ImageNet
Resolution: short-side 256
Training Data: Kinetics-400
Training Resources: 8 GPUs
Modality: RGB
Name: i3d_r50_lazy_32x2x1_100e_kinetics400_rgb
Results:
- Dataset: Kinetics-400
Metrics:
Top 1 Accuracy: 73.24
Top 5 Accuracy: 90.99
Task: Action Recognition
Training Json Log: https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_fast_256p_32x2x1_100e_kinetics400_rgb/20200725_031457.log.json
Training Log: https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_fast_256p_32x2x1_100e_kinetics400_rgb/20200725_031457.log
Weights: https://download.openmmlab.com/mmaction/recognition/i3d/i3d_r50_fast_256p_32x2x1_100e_kinetics400_rgb/i3d_r50_fast_256p_32x2x1_100e_kinetics400_rgb_20200817-4e90d1d5.pth
- Config: configs/recognition/i3d/i3d_nl_embedded_gaussian_r50_32x2x1_100e_kinetics400_rgb.py
In Collection: I3D
Metadata:
Architecture: ResNet50
Batch Size: 8
Epochs: 100
FLOPs: 54334488576
Parameters: 35397840
Pretrained: ImageNet
Resolution: short-side 256p
Training Data: Kinetics-400
Training Resources: 32 GPUs
Modality: RGB
Name: i3d_nl_embedded_gaussian_r50_32x2x1_100e_kinetics400_rgb
Results:
- Dataset: Kinetics-400
Metrics:
Top 1 Accuracy: 74.71
Top 5 Accuracy: 91.81
Task: Action Recognition
Training Json Log: https://download.openmmlab.com/mmaction/recognition/i3d/i3d_nl_embedded_gaussian_r50_32x2x1_100e_kinetics400_rgb/20200813_034054.log.json
Training Log: https://download.openmmlab.com/mmaction/recognition/i3d/i3d_nl_embedded_gaussian_r50_32x2x1_100e_kinetics400_rgb/20200813_034054.log
Weights: https://download.openmmlab.com/mmaction/recognition/i3d/i3d_nl_embedded_gaussian_r50_32x2x1_100e_kinetics400_rgb/i3d_nl_embedded_gaussian_r50_32x2x1_100e_kinetics400_rgb_20200813-6e6aef1b.pth
- Config: configs/recognition/i3d/i3d_nl_gaussian_r50_32x2x1_100e_kinetics400_rgb.py
In Collection: I3D
Metadata:
Architecture: ResNet50
Batch Size: 8
Epochs: 100
FLOPs: 48962109440
Parameters: 31723728
Pretrained: ImageNet
Resolution: short-side 256p
Training Data: Kinetics-400
Training Resources: 32 GPUs
Modality: RGB
Name: i3d_nl_gaussian_r50_32x2x1_100e_kinetics400_rgb
Results:
- Dataset: Kinetics-400
Metrics:
Top 1 Accuracy: 73.37
Top 5 Accuracy: 91.26
Task: Action Recognition
Training Json Log: https://download.openmmlab.com/mmaction/recognition/i3d/i3d_nl_gaussian_r50_32x2x1_100e_kinetics400_rgb/20200813_034909.log.json
Training Log: https://download.openmmlab.com/mmaction/recognition/i3d/i3d_nl_gaussian_r50_32x2x1_100e_kinetics400_rgb/20200813_034909.log
Weights: https://download.openmmlab.com/mmaction/recognition/i3d/i3d_nl_gaussian_r50_32x2x1_100e_kinetics400_rgb/i3d_nl_gaussian_r50_32x2x1_100e_kinetics400_rgb_20200815-17f84aa2.pth
- Config: configs/recognition/i3d/i3d_nl_dot_product_r50_32x2x1_100e_kinetics400_rgb.py
In Collection: I3D
Metadata:
Architecture: ResNet50
Batch Size: 8
Epochs: 100
FLOPs: 54334488576
Parameters: 35397840
Pretrained: ImageNet
Resolution: short-side 256p
Training Data: Kinetics-400
Training Resources: 32 GPUs
Modality: RGB
Name: i3d_nl_dot_product_r50_32x2x1_100e_kinetics400_rgb
Results:
- Dataset: Kinetics-400
Metrics:
Top 1 Accuracy: 73.92
Top 5 Accuracy: 91.59
Task: Action Recognition
Training Json Log: https://download.openmmlab.com/mmaction/recognition/i3d/i3d_nl_dot_product_r50_32x2x1_100e_kinetics400_rgb/20200814_044208.log.json
Training Log: https://download.openmmlab.com/mmaction/recognition/i3d/i3d_nl_dot_product_r50_32x2x1_100e_kinetics400_rgb/20200814_044208.log
Weights: https://download.openmmlab.com/mmaction/recognition/i3d/i3d_nl_dot_product_r50_32x2x1_100e_kinetics400_rgb/i3d_nl_dot_product_r50_32x2x1_100e_kinetics400_rgb_20200814-7c30d5bb.pth
configs/recognition/omnisource/README.md 0 → 100644
View file @ aa58d024
# Omni-sourced Webly-supervised Learning for Video Recognition
[Omni-sourced Webly-supervised Learning for Video Recognition](https://arxiv.org/abs/2003.13042)
[Dataset](https://docs.google.com/forms/d/e/1FAIpQLSd8_GlmHzG8FcDbW-OEu__G7qLgOSYZpH-i5vYVJcu7wcb_TQ/viewform?usp=sf_link)
## Abstract
<!-- [ABSTRACT] -->
We introduce OmniSource, a novel framework for leveraging web data to train video recognition models. OmniSource overcomes the barriers between data formats, such as images, short videos, and long untrimmed videos for webly-supervised learning. First, data samples with multiple formats, curated by task-specific data collection and automatically filtered by a teacher model, are transformed into a unified form. Then a joint-training strategy is proposed to deal with the domain gaps between multiple data sources and formats in webly-supervised learning. Several good practices, including data balancing, resampling, and cross-dataset mixup are adopted in joint training. Experiments show that by utilizing data from multiple sources and formats, OmniSource is more data-efficient in training. With only 3.5M images and 800K minutes videos crawled from the internet without human labeling (less than 2% of prior works), our models learned with OmniSource improve Top-1 accuracy of 2D- and 3D-ConvNet baseline models by 3.0% and 3.9%, respectively, on the Kinetics-400 benchmark. With OmniSource, we establish new records with different pretraining strategies for video recognition. Our best models achieve 80.4%, 80.5%, and 83.6 Top-1 accuracies on the Kinetics-400 benchmark respectively for training-from-scratch, ImageNet pre-training and IG-65M pre-training.
<!-- [IMAGE] -->
<div align=center>
<img src="https://github.com/open-mmlab/mmaction2/blob/master/configs/recognition/omnisource/pipeline.png" width="800"/>
</div>
## Results and Models
### Kinetics-400 Model Release
We currently released 4 models trained with OmniSource framework, including both 2D and 3D architectures. We compare the performance of models trained with or without OmniSource in the following table.
| Model | Modality | Pretrained | Backbone | Input | Resolution | Top-1 (Baseline / OmniSource (Delta)) | Top-5 (Baseline / OmniSource (Delta))) | Download |
| :------: | :------: | :--------: | :-------: | :---: | :------------: | :-----------------------------------: | :------------------------------------: | :-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: |
| TSN | RGB | ImageNet | ResNet50 | 3seg | 340x256 | 70.6 / 73.6 (+ 3.0) | 89.4 / 91.0 (+ 1.6) | [Baseline](https://download.openmmlab.com/mmaction/recognition/tsn/tsn_r50_1x1x3_100e_kinetics400_rgb/tsn_r50_1x1x3_100e_kinetics400_rgb_20200614-e508be42.pth) / [OmniSource](https://download.openmmlab.com/mmaction/recognition/tsn/omni/tsn_imagenet_pretrained_r50_omni_1x1x3_kinetics400_rgb_20200926-54192355.pth) |
| TSN | RGB | IG-1B | ResNet50 | 3seg | short-side 320 | 73.1 / 75.7 (+ 2.6) | 90.4 / 91.9 (+ 1.5) | [Baseline](https://download.openmmlab.com/mmaction/recognition/tsn/omni/tsn_1G1B_pretrained_r50_without_omni_1x1x3_kinetics400_rgb_20200926-c133dd49.pth) / [OmniSource](https://download.openmmlab.com/mmaction/recognition/tsn/omni/tsn_1G1B_pretrained_r50_omni_1x1x3_kinetics400_rgb_20200926-2863fed0.pth) |
| SlowOnly | RGB | Scratch | ResNet50 | 4x16 | short-side 320 | 72.9 / 76.8 (+ 3.9) | 90.9 / 92.5 (+ 1.6) | [Baseline](https://download.openmmlab.com/mmaction/recognition/slowonly/slowonly_r50_4x16x1_256e_kinetics400_rgb/slowonly_r50_4x16x1_256e_kinetics400_rgb_20200704-a69556c6.pth) / [OmniSource](https://download.openmmlab.com/mmaction/recognition/slowonly/omni/slowonly_r50_omni_4x16x1_kinetics400_rgb_20200926-51b1f7ea.pth) |
| SlowOnly | RGB | Scratch | ResNet101 | 8x8 | short-side 320 | 76.5 / 80.4 (+ 3.9) | 92.7 / 94.4 (+ 1.7) | [Baseline](https://download.openmmlab.com/mmaction/recognition/slowonly/omni/slowonly_r101_without_omni_8x8x1_kinetics400_rgb_20200926-0c730aef.pth) / [OmniSource](https://download.openmmlab.com/mmaction/recognition/slowonly/omni/slowonly_r101_omni_8x8x1_kinetics400_rgb_20200926-b5dbb701.pth) |
1. The validation set of Kinetics400 we used consists of 19796 videos. These videos are available at [Kinetics400-Validation](https://mycuhk-my.sharepoint.com/:u:/g/personal/1155136485_link_cuhk_edu_hk/EbXw2WX94J1Hunyt3MWNDJUBz-nHvQYhO9pvKqm6g39PMA?e=a9QldB). The corresponding [data list](https://download.openmmlab.com/mmaction/dataset/k400_val/kinetics_val_list.txt) (each line is of the format 'video_id, num_frames, label_index') and the [label map](https://download.openmmlab.com/mmaction/dataset/k400_val/kinetics_class2ind.txt) are also available.
## Benchmark on Mini-Kinetics
We release a subset of web dataset used in the OmniSource paper. Specifically, we release the web data in the 200 classes of [Mini-Kinetics](https://arxiv.org/pdf/1712.04851.pdf). The statistics of those datasets is detailed in [preparing_omnisource](/tools/data/omnisource/README.md). To obtain those data, you need to fill in a [data request form](https://docs.google.com/forms/d/e/1FAIpQLSd8_GlmHzG8FcDbW-OEu__G7qLgOSYZpH-i5vYVJcu7wcb_TQ/viewform?usp=sf_link). After we received your request, the download link of these data will be send to you. For more details on the released OmniSource web dataset, please refer to [preparing_omnisource](/tools/data/omnisource/README.md).
We benchmark the OmniSource framework on the released subset, results are listed in the following table (we report the Top-1 and Top-5 accuracy on Mini-Kinetics validation). The benchmark can be used as a baseline for video recognition with web data.
### TSN-8seg-ResNet50
| Model | Modality | Pretrained | Backbone | Input | Resolution | top1 acc | top5 acc | ckpt | json | log |
| :-------------------------------------------------------------------------------------------------------------------------------------------------------------------: | -------- | ---------- | -------- | ----- | -------------- | :------: | :------: | :------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: | :----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: | :--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: |
| [tsn_r50_1x1x8_100e_minikinetics_rgb](/configs/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics/tsn_r50_1x1x8_100e_minikinetics_rgb.py) | RGB | ImageNet | ResNet50 | 3seg | short-side 320 | 77.4 | 93.6 | [ckpt](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/baseline/tsn_r50_1x1x8_100e_minikinetics_rgb_20201030-b4eaf92b.pth) | [json](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/baseline/tsn_r50_1x1x8_100e_minikinetics_rgb_20201030.json) | [log](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/baseline/tsn_r50_1x1x8_100e_minikinetics_rgb_20201030.log) |
| [tsn_r50_1x1x8_100e_minikinetics_googleimage_rgb](/configs/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics/tsn_r50_1x1x8_100e_minikinetics_googleimage_rgb.py) | RGB | ImageNet | ResNet50 | 3seg | short-side 320 | 78.0 | 93.6 | [ckpt](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/googleimage/tsn_r50_1x1x8_100e_minikinetics_googleimage_rgb_20201030-23966b4b.pth) | [json](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/googleimage/tsn_r50_1x1x8_100e_minikinetics_googleimage_rgb_20201030.json) | [log](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/googleimage/tsn_r50_1x1x8_100e_minikinetics_googleimage_rgb_20201030.log) |
| [tsn_r50_1x1x8_100e_minikinetics_webimage_rgb](/configs/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics/tsn_r50_1x1x8_100e_minikinetics_webimage_rgb.py) | RGB | ImageNet | ResNet50 | 3seg | short-side 320 | 78.6 | 93.6 | [ckpt](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/webimage/tsn_r50_1x1x8_100e_minikinetics_webimage_rgb_20201030-66f5e046.pth) | [json](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/webimage/tsn_r50_1x1x8_100e_minikinetics_webimage_rgb_20201030.json) | [log](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/webimage/tsn_r50_1x1x8_100e_minikinetics_webimage_rgb_20201030.log) |
| [tsn_r50_1x1x8_100e_minikinetics_insvideo_rgb](/configs/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics/tsn_r50_1x1x8_100e_minikinetics_insvideo_rgb.py) | RGB | ImageNet | ResNet50 | 3seg | short-side 320 | 80.6 | 95.0 | [ckpt](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/insvideo/tsn_r50_1x1x8_100e_minikinetics_insvideo_rgb_20201030-011f984d.pth) | [json](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/insvideo/tsn_r50_1x1x8_100e_minikinetics_insvideo_rgb_20201030.json) | [log](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/insvideo/tsn_r50_1x1x8_100e_minikinetics_insvideo_rgb_20201030.log) |
| [tsn_r50_1x1x8_100e_minikinetics_kineticsraw_rgb](/configs/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics/tsn_r50_1x1x8_100e_minikinetics_kineticsraw_rgb.py) | RGB | ImageNet | ResNet50 | 3seg | short-side 320 | 78.6 | 93.2 | [ckpt](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/kineticsraw/tsn_r50_1x1x8_100e_minikinetics_kineticsraw_rgb_20201030-59f5d064.pth) | [json](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/kineticsraw/tsn_r50_1x1x8_100e_minikinetics_kineticsraw_rgb_20201030.json) | [log](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/kineticsraw/tsn_r50_1x1x8_100e_minikinetics_kineticsraw_rgb_20201030.log) |
| [tsn_r50_1x1x8_100e_minikinetics_omnisource_rgb](/configs/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics/tsn_r50_1x1x8_100e_minikinetics_omnisource_rgb.py) | RGB | ImageNet | ResNet50 | 3seg | short-side 320 | 81.3 | 94.8 | [ckpt](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/omnisource/tsn_r50_1x1x8_100e_minikinetics_omnisource_rgb_20201030-0f56ef51.pth) | [json](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/omnisource/tsn_r50_1x1x8_100e_minikinetics_omnisource_rgb_20201030.json) | [log](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/omnisource/tsn_r50_1x1x8_100e_minikinetics_omnisource_rgb_20201030.log) |
### SlowOnly-8x8-ResNet50
| Model | Modality | Pretrained | Backbone | Input | Resolution | top1 acc | top5 acc | ckpt | json | log |
| :----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: | -------- | ---------- | -------- | ----- | -------------- | :------: | :------: | :----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: | :--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: | :------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: |
| [slowonly_r50_8x8x1_256e_minikinetics_rgb](/configs/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics/slowonly_r50_8x8x1_256e_minikinetics_rgb.py) | RGB | None | ResNet50 | 8x8 | short-side 320 | 78.6 | 93.9 | [ckpt](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/baseline/slowonly_r50_8x8x1_256e_minikinetics_rgb_20201030-168eb098.pth) | [json](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/baseline/slowonly_r50_8x8x1_256e_minikinetics_rgb_20201030.json) | [log](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/baseline/slowonly_r50_8x8x1_256e_minikinetics_rgb_20201030.log) |
| [slowonly_r50_8x8x1_256e_minikinetics_googleimage_rgb](/configs/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics/slowonly_r50_8x8x1_256e_minikinetics_googleimage_rgb.py) | RGB | None | ResNet50 | 8x8 | short-side 320 | 80.8 | 95.0 | [ckpt](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/googleimage/slowonly_r50_8x8x1_256e_minikinetics_googleimage_rgb_20201030-7da6dfc3.pth) | [json](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/googleimage/slowonly_r50_8x8x1_256e_minikinetics_googleimage_rgb_20201030.json) | [log](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/googleimage/slowonly_r50_8x8x1_256e_minikinetics_googleimage_rgb_20201030.log) |
| [slowonly_r50_8x8x1_256e_minikinetics_webimage_rgb](/configs/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics/slowonly_r50_8x8x1_256e_minikinetics_webimage_rgb.py) | RGB | None | ResNet50 | 8x8 | short-side 320 | 81.3 | 95.2 | [ckpt](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/webimage/slowonly_r50_8x8x1_256e_minikinetics_webimage_rgb_20201030-c36616e9.pth) | [json](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/webimage/slowonly_r50_8x8x1_256e_minikinetics_webimage_rgb_20201030.json) | [log](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/webimage/slowonly_r50_8x8x1_256e_minikinetics_webimage_rgb_20201030.log) |
| [slowonly_r50_8x8x1_256e_minikinetics_insvideo_rgb](/configs/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics/slowonly_r50_8x8x1_256e_minikinetics_insvideo_rgb.py) | RGB | None | ResNet50 | 8x8 | short-side 320 | 82.4 | 95.6 | [ckpt](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/insvideo/slowonly_r50_8x8x1_256e_minikinetics_insvideo_rgb_20201030-e2890e8d.pth) | [json](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/insvideo/slowonly_r50_8x8x1_256e_minikinetics_insvideo_rgb_20201030.json) | [log](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/insvideo/slowonly_r50_8x8x1_256e_minikinetics_insvideo_rgb_20201030.log) |
| [slowonly_r50_8x8x1_256e_minikinetics_kineticsraw_rgb](/configs/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics/slowonly_r50_8x8x1_256e_minikinetics_kineticsraw_rgb.py) | RGB | None | ResNet50 | 8x8 | short-side 320 | 80.3 | 94.5 | [ckpt](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/kineticsraw/slowonly_r50_8x8x1_256e_minikinetics_kineticsraw_rgb_20201030-62974bac.pth) | [json](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/kineticsraw/slowonly_r50_8x8x1_256e_minikinetics_kineticsraw_rgb_20201030.json) | [log](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/kineticsraw/slowonly_r50_8x8x1_256e_minikinetics_kineticsraw_rgb_20201030.log) |
| [slowonly_r50_8x8x1_256e_minikinetics_omnisource_rgb](/configs/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics/slowonly_r50_8x8x1_256e_minikinetics_googleimage_rgb.py) | RGB | None | ResNet50 | 8x8 | short-side 320 | 82.9 | 95.8 | [ckpt](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/omnisource/slowonly_r50_8x8x1_256e_minikinetics_omnisource_rgb_20201030-284cfd3b.pth) | [json](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/omnisource/slowonly_r50_8x8x1_256e_minikinetics_omnisource_rgb_20201030.json) | [log](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/omnisource/slowonly_r50_8x8x1_256e_minikinetics_omnisource_rgb_20201030.log) |
We also list the benchmark in the original paper which run on Kinetics-400 for comparison:
| Model | Baseline | +GG-img | +\[GG-IG\]-img | +IG-vid | +KRaw | OmniSource |
| :--------------------: | :---------: | :---------: | :------------: | :---------: | :---------: | :---------: |
| TSN-3seg-ResNet50 | 70.6 / 89.4 | 71.5 / 89.5 | 72.0 / 90.0 | 72.0 / 90.3 | 71.7 / 89.6 | 73.6 / 91.0 |
| SlowOnly-4x16-ResNet50 | 73.8 / 90.9 | 74.5 / 91.4 | 75.2 / 91.6 | 75.2 / 91.7 | 74.5 / 91.1 | 76.6 / 92.5 |
## Citation
<!-- [ALGORITHM] -->
```BibTeX
@article{duan2020omni,
title={Omni-sourced Webly-supervised Learning for Video Recognition},
author={Duan, Haodong and Zhao, Yue and Xiong, Yuanjun and Liu, Wentao and Lin, Dahua},
journal={arXiv preprint arXiv:2003.13042},
year={2020}
}
```
configs/recognition/omnisource/README_zh-CN.md 0 → 100644
View file @ aa58d024
# Omni-sourced Webly-supervised Learning for Video Recognition
[Haodong Duan](https://github.com/kennymckormick), [Yue Zhao](https://github.com/zhaoyue-zephyrus), [Yuanjun Xiong](https://github.com/yjxiong), Wentao Liu, [Dahua Lin](https://github.com/lindahua)
In ECCV, 2020. [Paper](https://arxiv.org/abs/2003.13042), [Dataset](https://docs.google.com/forms/d/e/1FAIpQLSd8_GlmHzG8FcDbW-OEu__G7qLgOSYZpH-i5vYVJcu7wcb_TQ/viewform?usp=sf_link)
![pipeline](https://github.com/open-mmlab/mmaction2/blob/master/configs/recognition/omnisource/pipeline.png?raw=true)
## 模型库
### Kinetics-400
MMAction2 当前公开了 4 个 OmniSource 框架训练的模型，包含 2D 架构与 3D 架构。下表比较了使用或不适用 OmniSource 框架训练得的模型在 Kinetics-400 上的精度：
| 模型 | 模态 | 预训练 | 主干网络 | 输入 | 分辨率 | Top-1 准确率(Baseline / OmniSource (Delta)) | Top-5 准确率(Baseline / OmniSource (Delta))) | 模型下载链接 |
| :------: | :--: | :------: | :-------: | :--: | :------------: | :-----------------------------------------: | :------------------------------------------: | :-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: |
| TSN | RGB | ImageNet | ResNet50 | 3seg | 340x256 | 70.6 / 73.6 (+ 3.0) | 89.4 / 91.0 (+ 1.6) | [Baseline](https://download.openmmlab.com/mmaction/recognition/tsn/tsn_r50_1x1x3_100e_kinetics400_rgb/tsn_r50_1x1x3_100e_kinetics400_rgb_20200614-e508be42.pth) / [OmniSource](https://download.openmmlab.com/mmaction/recognition/tsn/omni/tsn_imagenet_pretrained_r50_omni_1x1x3_kinetics400_rgb_20200926-54192355.pth) |
| TSN | RGB | IG-1B | ResNet50 | 3seg | short-side 320 | 73.1 / 75.7 (+ 2.6) | 90.4 / 91.9 (+ 1.5) | [Baseline](https://download.openmmlab.com/mmaction/recognition/tsn/omni/tsn_1G1B_pretrained_r50_without_omni_1x1x3_kinetics400_rgb_20200926-c133dd49.pth) / [OmniSource](https://download.openmmlab.com/mmaction/recognition/tsn/omni/tsn_1G1B_pretrained_r50_omni_1x1x3_kinetics400_rgb_20200926-2863fed0.pth) |
| SlowOnly | RGB | None | ResNet50 | 4x16 | short-side 320 | 72.9 / 76.8 (+ 3.9) | 90.9 / 92.5 (+ 1.6) | [Baseline](https://download.openmmlab.com/mmaction/recognition/slowonly/slowonly_r50_4x16x1_256e_kinetics400_rgb/slowonly_r50_4x16x1_256e_kinetics400_rgb_20200704-a69556c6.pth) / [OmniSource](https://download.openmmlab.com/mmaction/recognition/slowonly/omni/slowonly_r50_omni_4x16x1_kinetics400_rgb_20200926-51b1f7ea.pth) |
| SlowOnly | RGB | None | ResNet101 | 8x8 | short-side 320 | 76.5 / 80.4 (+ 3.9) | 92.7 / 94.4 (+ 1.7) | [Baseline](https://download.openmmlab.com/mmaction/recognition/slowonly/omni/slowonly_r101_without_omni_8x8x1_kinetics400_rgb_20200926-0c730aef.pth) / [OmniSource](https://download.openmmlab.com/mmaction/recognition/slowonly/omni/slowonly_r101_omni_8x8x1_kinetics400_rgb_20200926-b5dbb701.pth) |
1. 我们使用的 Kinetics400 验证集包含 19796 个视频，用户可以从 [验证集视频](https://mycuhk-my.sharepoint.com/:u:/g/personal/1155136485_link_cuhk_edu_hk/EbXw2WX94J1Hunyt3MWNDJUBz-nHvQYhO9pvKqm6g39PMA?e=a9QldB) 下载这些视频。同时也提供了对应的 [数据列表](https://download.openmmlab.com/mmaction/dataset/k400_val/kinetics_val_list.txt) （每行格式为：视频 ID，视频帧数目，类别序号）以及 [标签映射](https://download.openmmlab.com/mmaction/dataset/k400_val/kinetics_class2ind.txt) （类别序号到类别名称）。
## Mini-Kinetics 上的基准测试
OmniSource 项目当前公开了所采集网络数据的一个子集，涉及 [Mini-Kinetics](https://arxiv.org/pdf/1712.04851.pdf) 中的 200 个动作类别。[OmniSource 数据集准备](/tools/data/omnisource/README_zh-CN.md) 中记录了这些数据集的详细统计信息。用户可以通过填写 [申请表](https://docs.google.com/forms/d/e/1FAIpQLSd8_GlmHzG8FcDbW-OEu__G7qLgOSYZpH-i5vYVJcu7wcb_TQ/viewform?usp=sf_link) 获取这些数据，在完成填写后，数据下载链接会被发送至用户邮箱。更多关于 OmniSource 网络数据集的信息请参照 [OmniSource 数据集准备](/tools/data/omnisource/README_zh-CN.md)。
MMAction2 在公开的数据集上进行了 OmniSource 框架的基准测试，下表记录了详细的结果（在 Mini-Kinetics 验证集上的精度），这些结果可以作为使用网络数据训练视频识别任务的基线。
### TSN-8seg-ResNet50
| 模型 | 模态 | 预训练 | 主干网络 | 输入 | 分辨率 | Top-1 准确率 | Top-5 准确率 | ckpt | json | log |
| :-------------------------------------------------------------------------------------------------------------------------------------------------------------------: | :--: | :------: | :------: | :--: | :------------: | :----------: | :----------: | :------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: | :----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: | :--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: |
| [tsn_r50_1x1x8_100e_minikinetics_rgb](/configs/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics/tsn_r50_1x1x8_100e_minikinetics_rgb.py) | RGB | ImageNet | ResNet50 | 3seg | short-side 320 | 77.4 | 93.6 | [ckpt](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/baseline/tsn_r50_1x1x8_100e_minikinetics_rgb_20201030-b4eaf92b.pth) | [json](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/baseline/tsn_r50_1x1x8_100e_minikinetics_rgb_20201030.json) | [log](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/baseline/tsn_r50_1x1x8_100e_minikinetics_rgb_20201030.log) |
| [tsn_r50_1x1x8_100e_minikinetics_googleimage_rgb](/configs/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics/tsn_r50_1x1x8_100e_minikinetics_googleimage_rgb.py) | RGB | ImageNet | ResNet50 | 3seg | short-side 320 | 78.0 | 93.6 | [ckpt](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/googleimage/tsn_r50_1x1x8_100e_minikinetics_googleimage_rgb_20201030-23966b4b.pth) | [json](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/googleimage/tsn_r50_1x1x8_100e_minikinetics_googleimage_rgb_20201030.json) | [log](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/googleimage/tsn_r50_1x1x8_100e_minikinetics_googleimage_rgb_20201030.log) |
| [tsn_r50_1x1x8_100e_minikinetics_webimage_rgb](/configs/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics/tsn_r50_1x1x8_100e_minikinetics_webimage_rgb.py) | RGB | ImageNet | ResNet50 | 3seg | short-side 320 | 78.6 | 93.6 | [ckpt](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/webimage/tsn_r50_1x1x8_100e_minikinetics_webimage_rgb_20201030-66f5e046.pth) | [json](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/webimage/tsn_r50_1x1x8_100e_minikinetics_webimage_rgb_20201030.json) | [log](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/webimage/tsn_r50_1x1x8_100e_minikinetics_webimage_rgb_20201030.log) |
| [tsn_r50_1x1x8_100e_minikinetics_insvideo_rgb](/configs/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics/tsn_r50_1x1x8_100e_minikinetics_insvideo_rgb.py) | RGB | ImageNet | ResNet50 | 3seg | short-side 320 | 80.6 | 95.0 | [ckpt](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/insvideo/tsn_r50_1x1x8_100e_minikinetics_insvideo_rgb_20201030-011f984d.pth) | [json](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/insvideo/tsn_r50_1x1x8_100e_minikinetics_insvideo_rgb_20201030.json) | [log](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/insvideo/tsn_r50_1x1x8_100e_minikinetics_insvideo_rgb_20201030.log) |
| [tsn_r50_1x1x8_100e_minikinetics_kineticsraw_rgb](/configs/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics/tsn_r50_1x1x8_100e_minikinetics_kineticsraw_rgb.py) | RGB | ImageNet | ResNet50 | 3seg | short-side 320 | 78.6 | 93.2 | [ckpt](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/kineticsraw/tsn_r50_1x1x8_100e_minikinetics_kineticsraw_rgb_20201030-59f5d064.pth) | [json](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/kineticsraw/tsn_r50_1x1x8_100e_minikinetics_kineticsraw_rgb_20201030.json) | [log](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/kineticsraw/tsn_r50_1x1x8_100e_minikinetics_kineticsraw_rgb_20201030.log) |
| [tsn_r50_1x1x8_100e_minikinetics_omnisource_rgb](/configs/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics/tsn_r50_1x1x8_100e_minikinetics_omnisource_rgb.py) | RGB | ImageNet | ResNet50 | 3seg | short-side 320 | 81.3 | 94.8 | [ckpt](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/omnisource/tsn_r50_1x1x8_100e_minikinetics_omnisource_rgb_20201030-0f56ef51.pth) | [json](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/omnisource/tsn_r50_1x1x8_100e_minikinetics_omnisource_rgb_20201030.json) | [log](https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/omnisource/tsn_r50_1x1x8_100e_minikinetics_omnisource_rgb_20201030.log) |
### SlowOnly-8x8-ResNet50
| 模型 | 模态 | 预训练 | 主干网络 | 输入 | 分辨率 | Top-1 准确率 | Top-5 准确率 | ckpt | json | log |
| :----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: | :--: | :----: | :------: | :--: | :------------: | :----------: | :----------: | :----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: | :--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: | :------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: |
| [slowonly_r50_8x8x1_256e_minikinetics_rgb](/configs/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics/slowonly_r50_8x8x1_256e_minikinetics_rgb.py) | RGB | None | ResNet50 | 8x8 | short-side 320 | 78.6 | 93.9 | [ckpt](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/baseline/slowonly_r50_8x8x1_256e_minikinetics_rgb_20201030-168eb098.pth) | [json](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/baseline/slowonly_r50_8x8x1_256e_minikinetics_rgb_20201030.json) | [log](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/baseline/slowonly_r50_8x8x1_256e_minikinetics_rgb_20201030.log) |
| [slowonly_r50_8x8x1_256e_minikinetics_googleimage_rgb](/configs/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics/slowonly_r50_8x8x1_256e_minikinetics_googleimage_rgb.py) | RGB | None | ResNet50 | 8x8 | short-side 320 | 80.8 | 95.0 | [ckpt](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/googleimage/slowonly_r50_8x8x1_256e_minikinetics_googleimage_rgb_20201030-7da6dfc3.pth) | [json](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/googleimage/slowonly_r50_8x8x1_256e_minikinetics_googleimage_rgb_20201030.json) | [log](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/googleimage/slowonly_r50_8x8x1_256e_minikinetics_googleimage_rgb_20201030.log) |
| [slowonly_r50_8x8x1_256e_minikinetics_webimage_rgb](/configs/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics/slowonly_r50_8x8x1_256e_minikinetics_webimage_rgb.py) | RGB | None | ResNet50 | 8x8 | short-side 320 | 81.3 | 95.2 | [ckpt](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/webimage/slowonly_r50_8x8x1_256e_minikinetics_webimage_rgb_20201030-c36616e9.pth) | [json](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/webimage/slowonly_r50_8x8x1_256e_minikinetics_webimage_rgb_20201030.json) | [log](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/webimage/slowonly_r50_8x8x1_256e_minikinetics_webimage_rgb_20201030.log) |
| [slowonly_r50_8x8x1_256e_minikinetics_insvideo_rgb](/configs/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics/slowonly_r50_8x8x1_256e_minikinetics_insvideo_rgb.py) | RGB | None | ResNet50 | 8x8 | short-side 320 | 82.4 | 95.6 | [ckpt](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/insvideo/slowonly_r50_8x8x1_256e_minikinetics_insvideo_rgb_20201030-e2890e8d.pth) | [json](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/insvideo/slowonly_r50_8x8x1_256e_minikinetics_insvideo_rgb_20201030.json) | [log](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/insvideo/slowonly_r50_8x8x1_256e_minikinetics_insvideo_rgb_20201030.log) |
| [slowonly_r50_8x8x1_256e_minikinetics_kineticsraw_rgb](/configs/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics/slowonly_r50_8x8x1_256e_minikinetics_kineticsraw_rgb.py) | RGB | None | ResNet50 | 8x8 | short-side 320 | 80.3 | 94.5 | [ckpt](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/kineticsraw/slowonly_r50_8x8x1_256e_minikinetics_kineticsraw_rgb_20201030-62974bac.pth) | [json](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/kineticsraw/slowonly_r50_8x8x1_256e_minikinetics_kineticsraw_rgb_20201030.json) | [log](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/kineticsraw/slowonly_r50_8x8x1_256e_minikinetics_kineticsraw_rgb_20201030.log) |
| [slowonly_r50_8x8x1_256e_minikinetics_omnisource_rgb](/configs/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics/slowonly_r50_8x8x1_256e_minikinetics_googleimage_rgb.py) | RGB | None | ResNet50 | 8x8 | short-side 320 | 82.9 | 95.8 | [ckpt](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/omnisource/slowonly_r50_8x8x1_256e_minikinetics_omnisource_rgb_20201030-284cfd3b.pth) | [json](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/omnisource/slowonly_r50_8x8x1_256e_minikinetics_omnisource_rgb_20201030.json) | [log](https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/omnisource/slowonly_r50_8x8x1_256e_minikinetics_omnisource_rgb_20201030.log) |
下表列出了原论文中在 Kinetics-400 上进行基准测试的结果供参考：
| Model | Baseline | +GG-img | +\[GG-IG\]-img | +IG-vid | +KRaw | OmniSource |
| :--------------------: | :---------: | :---------: | :------------: | :---------: | :---------: | :---------: |
| TSN-3seg-ResNet50 | 70.6 / 89.4 | 71.5 / 89.5 | 72.0 / 90.0 | 72.0 / 90.3 | 71.7 / 89.6 | 73.6 / 91.0 |
| SlowOnly-4x16-ResNet50 | 73.8 / 90.9 | 74.5 / 91.4 | 75.2 / 91.6 | 75.2 / 91.7 | 74.5 / 91.1 | 76.6 / 92.5 |
## 注：
如果 OmniSource 项目对您的研究有所帮助，请使用以下 BibTex 项进行引用：
<!-- [ALGORITHM] -->
```BibTeX
@article{duan2020omni,
title={Omni-sourced Webly-supervised Learning for Video Recognition},
author={Duan, Haodong and Zhao, Yue and Xiong, Yuanjun and Liu, Wentao and Lin, Dahua},
journal={arXiv preprint arXiv:2003.13042},
year={2020}
}
```
configs/recognition/omnisource/metafile.yml 0 → 100644
View file @ aa58d024
Collections:
- Name: OmniSource
README: configs/recognition/omnisource/README.md
Paper:
URL: https://arxiv.org/abs/2003.13042
Title: Omni-sourced Webly-supervised Learning for Video Recognition
Models:
- Config: configs/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics/tsn_r50_1x1x8_100e_minikinetics_rgb.py
In Collection: OmniSource
Metadata:
Architecture: ResNet50
Batch Size: 12
Epochs: 100
FLOPs: 134526976000
Input: 3seg
Modality: RGB
Parameters: 23917832
Pretrained: ImageNet
Resolution: short-side 320
Training Data: MiniKinetics
Modality: RGB
Name: tsn_r50_1x1x8_100e_minikinetics_rgb
Results:
- Dataset: MiniKinetics
Metrics:
Top 1 Accuracy: 77.4
Top 5 Accuracy: 93.6
Task: Action Recognition
Training Json Log: https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/baseline/tsn_r50_1x1x8_100e_minikinetics_rgb_20201030.json
Training Log: https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/baseline/tsn_r50_1x1x8_100e_minikinetics_rgb_20201030.log
Weights: https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/baseline/tsn_r50_1x1x8_100e_minikinetics_rgb_20201030-b4eaf92b.pth
- Config: configs/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics/tsn_r50_1x1x8_100e_minikinetics_googleimage_rgb.py
In Collection: OmniSource
Metadata:
Architecture: ResNet50
Batch Size: 12
Epochs: 100
FLOPs: 134526976000
Input: 3seg
Modality: RGB
Parameters: 23917832
Pretrained: ImageNet
Resolution: short-side 320
Training Data: MiniKinetics
Modality: RGB
Name: tsn_r50_1x1x8_100e_minikinetics_googleimage_rgb
Results:
- Dataset: MiniKinetics
Metrics:
Top 1 Accuracy: 78.0
Top 5 Accuracy: 93.6
Task: Action Recognition
Training Json Log: https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/googleimage/tsn_r50_1x1x8_100e_minikinetics_googleimage_rgb_20201030.json
Training Log: https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/googleimage/tsn_r50_1x1x8_100e_minikinetics_googleimage_rgb_20201030.log
Weights: https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/googleimage/tsn_r50_1x1x8_100e_minikinetics_googleimage_rgb_20201030-23966b4b.pth
- Config: configs/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics/tsn_r50_1x1x8_100e_minikinetics_webimage_rgb.py
In Collection: OmniSource
Metadata:
Architecture: ResNet50
Batch Size: 12
Epochs: 100
FLOPs: 134526976000
Input: 3seg
Modality: RGB
Parameters: 23917832
Pretrained: ImageNet
Resolution: short-side 320
Training Data: MiniKinetics
Modality: RGB
Name: tsn_r50_1x1x8_100e_minikinetics_webimage_rgb
Results:
- Dataset: MiniKinetics
Metrics:
Top 1 Accuracy: 78.6
Top 5 Accuracy: 93.6
Task: Action Recognition
Training Json Log: https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/webimage/tsn_r50_1x1x8_100e_minikinetics_webimage_rgb_20201030.json
Training Log: https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/webimage/tsn_r50_1x1x8_100e_minikinetics_webimage_rgb_20201030.log
Weights: https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/webimage/tsn_r50_1x1x8_100e_minikinetics_webimage_rgb_20201030-66f5e046.pth
- Config: configs/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics/tsn_r50_1x1x8_100e_minikinetics_insvideo_rgb.py
In Collection: OmniSource
Metadata:
Architecture: ResNet50
Batch Size: 12
Epochs: 100
FLOPs: 134526976000
Input: 3seg
Modality: RGB
Parameters: 23917832
Pretrained: ImageNet
Resolution: short-side 320
Training Data: MiniKinetics
Modality: RGB
Name: tsn_r50_1x1x8_100e_minikinetics_insvideo_rgb
Results:
- Dataset: MiniKinetics
Metrics:
Top 1 Accuracy: 80.6
Top 5 Accuracy: 95.0
Task: Action Recognition
Training Json Log: https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/insvideo/tsn_r50_1x1x8_100e_minikinetics_insvideo_rgb_20201030.json
Training Log: https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/insvideo/tsn_r50_1x1x8_100e_minikinetics_insvideo_rgb_20201030.log
Weights: https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/insvideo/tsn_r50_1x1x8_100e_minikinetics_insvideo_rgb_20201030-011f984d.pth
- Config: configs/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics/tsn_r50_1x1x8_100e_minikinetics_kineticsraw_rgb.py
In Collection: OmniSource
Metadata:
Architecture: ResNet50
Batch Size: 12
Epochs: 100
FLOPs: 134526976000
Input: 3seg
Modality: RGB
Parameters: 23917832
Pretrained: ImageNet
Resolution: short-side 320
Training Data: MiniKinetics
Modality: RGB
Name: tsn_r50_1x1x8_100e_minikinetics_kineticsraw_rgb
Results:
- Dataset: MiniKinetics
Metrics:
Top 1 Accuracy: 78.6
Top 5 Accuracy: 93.2
Task: Action Recognition
Training Json Log: https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/kineticsraw/tsn_r50_1x1x8_100e_minikinetics_kineticsraw_rgb_20201030.json
Training Log: https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/kineticsraw/tsn_r50_1x1x8_100e_minikinetics_kineticsraw_rgb_20201030.log
Weights: https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/kineticsraw/tsn_r50_1x1x8_100e_minikinetics_kineticsraw_rgb_20201030-59f5d064.pth
- Config: configs/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics/tsn_r50_1x1x8_100e_minikinetics_omnisource_rgb.py
In Collection: OmniSource
Metadata:
Architecture: ResNet50
Batch Size: 12
Epochs: 100
FLOPs: 134526976000
Input: 3seg
Modality: RGB
Parameters: 23917832
Pretrained: ImageNet
Resolution: short-side 320
Training Data: MiniKinetics
Modality: RGB
Name: tsn_r50_1x1x8_100e_minikinetics_omnisource_rgb
Results:
- Dataset: MiniKinetics
Metrics:
Top 1 Accuracy: 81.3
Top 5 Accuracy: 94.8
Task: Action Recognition
Training Json Log: https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/omnisource/tsn_r50_1x1x8_100e_minikinetics_omnisource_rgb_20201030.json
Training Log: https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/omnisource/tsn_r50_1x1x8_100e_minikinetics_omnisource_rgb_20201030.log
Weights: https://download.openmmlab.com/mmaction/recognition/omnisource/tsn_r50_1x1x8_100e_minikinetics_rgb/omnisource/tsn_r50_1x1x8_100e_minikinetics_omnisource_rgb_20201030-0f56ef51.pth
- Config: configs/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics/slowonly_r50_8x8x1_256e_minikinetics_rgb.py
In Collection: OmniSource
Metadata:
Architecture: ResNet50
Batch Size: 12
Epochs: 256
FLOPs: 54860070912
Input: 8x8
Modality: RGB
Parameters: 32044296
Pretrained: None
Resolution: short-side 320
Training Data: MiniKinetics
Modality: RGB
Name: slowonly_r50_8x8x1_256e_minikinetics_rgb
Results:
- Dataset: MiniKinetics
Metrics:
Top 1 Accuracy: 78.6
Top 5 Accuracy: 93.9
Task: Action Recognition
Training Json Log: https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/baseline/slowonly_r50_8x8x1_256e_minikinetics_rgb_20201030.json
Training Log: https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/baseline/slowonly_r50_8x8x1_256e_minikinetics_rgb_20201030.log
Weights: https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/baseline/slowonly_r50_8x8x1_256e_minikinetics_rgb_20201030-168eb098.pth
- Config: configs/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics/slowonly_r50_8x8x1_256e_minikinetics_googleimage_rgb.py
In Collection: OmniSource
Metadata:
Architecture: ResNet50
Batch Size: 12
Epochs: 256
FLOPs: 54860070912
Input: 8x8
Modality: RGB
Parameters: 32044296
Pretrained: None
Resolution: short-side 320
Training Data: MiniKinetics
Modality: RGB
Name: slowonly_r50_8x8x1_256e_minikinetics_googleimage_rgb
Results:
- Dataset: MiniKinetics
Metrics:
Top 1 Accuracy: 80.8
Top 5 Accuracy: 95.0
Task: Action Recognition
Training Json Log: https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/googleimage/slowonly_r50_8x8x1_256e_minikinetics_googleimage_rgb_20201030.json
Training Log: https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/googleimage/slowonly_r50_8x8x1_256e_minikinetics_googleimage_rgb_20201030.log
Weights: https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/googleimage/slowonly_r50_8x8x1_256e_minikinetics_googleimage_rgb_20201030-7da6dfc3.pth
- Config: configs/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics/slowonly_r50_8x8x1_256e_minikinetics_webimage_rgb.py
In Collection: OmniSource
Metadata:
Architecture: ResNet50
Batch Size: 12
Epochs: 256
FLOPs: 54860070912
Input: 8x8
Modality: RGB
Parameters: 32044296
Pretrained: None
Resolution: short-side 320
Training Data: MiniKinetics
Modality: RGB
Name: slowonly_r50_8x8x1_256e_minikinetics_webimage_rgb
Results:
- Dataset: MiniKinetics
Metrics:
Top 1 Accuracy: 81.3
Top 5 Accuracy: 95.2
Task: Action Recognition
Training Json Log: https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/webimage/slowonly_r50_8x8x1_256e_minikinetics_webimage_rgb_20201030.json
Training Log: https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/webimage/slowonly_r50_8x8x1_256e_minikinetics_webimage_rgb_20201030.log
Weights: https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/webimage/slowonly_r50_8x8x1_256e_minikinetics_webimage_rgb_20201030-c36616e9.pth
- Config: configs/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics/slowonly_r50_8x8x1_256e_minikinetics_insvideo_rgb.py
In Collection: OmniSource
Metadata:
Architecture: ResNet50
Batch Size: 12
Epochs: 256
FLOPs: 54860070912
Input: 8x8
Modality: RGB
Parameters: 32044296
Pretrained: None
Resolution: short-side 320
Training Data: MiniKinetics
Modality: RGB
Name: slowonly_r50_8x8x1_256e_minikinetics_insvideo_rgb
Results:
- Dataset: MiniKinetics
Metrics:
Top 1 Accuracy: 82.4
Top 5 Accuracy: 95.6
Task: Action Recognition
Training Json Log: https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/insvideo/slowonly_r50_8x8x1_256e_minikinetics_insvideo_rgb_20201030.json
Training Log: https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/insvideo/slowonly_r50_8x8x1_256e_minikinetics_insvideo_rgb_20201030.log
Weights: https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/insvideo/slowonly_r50_8x8x1_256e_minikinetics_insvideo_rgb_20201030-e2890e8d.pth
- Config: configs/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics/slowonly_r50_8x8x1_256e_minikinetics_kineticsraw_rgb.py
In Collection: OmniSource
Metadata:
Architecture: ResNet50
Batch Size: 12
Epochs: 256
FLOPs: 54860070912
Input: 8x8
Modality: RGB
Parameters: 32044296
Pretrained: None
Resolution: short-side 320
Training Data: MiniKinetics
Modality: RGB
Name: slowonly_r50_8x8x1_256e_minikinetics_kineticsraw_rgb
Results:
- Dataset: MiniKinetics
Metrics:
Top 1 Accuracy: 80.3
Top 5 Accuracy: 94.5
Task: Action Recognition
Training Json Log: https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/kineticsraw/slowonly_r50_8x8x1_256e_minikinetics_kineticsraw_rgb_20201030.json
Training Log: https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/kineticsraw/slowonly_r50_8x8x1_256e_minikinetics_kineticsraw_rgb_20201030.log
Weights: https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/kineticsraw/slowonly_r50_8x8x1_256e_minikinetics_kineticsraw_rgb_20201030-62974bac.pth
- Config: configs/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics/slowonly_r50_8x8x1_256e_minikinetics_googleimage_rgb.py
In Collection: OmniSource
Metadata:
Architecture: ResNet50
Batch Size: 12
Epochs: 256
FLOPs: 54860070912
Input: 8x8
Modality: RGB
Parameters: 32044296
Pretrained: None
Resolution: short-side 320
Training Data: MiniKinetics
Modality: RGB
Name: slowonly_r50_8x8x1_256e_minikinetics_omnisource_rgb
Results:
- Dataset: MiniKinetics
Metrics:
Top 1 Accuracy: 82.9
Top 5 Accuracy: 95.8
Task: Action Recognition
Training Json Log: https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/omnisource/slowonly_r50_8x8x1_256e_minikinetics_omnisource_rgb_20201030.json
Training Log: https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/omnisource/slowonly_r50_8x8x1_256e_minikinetics_omnisource_rgb_20201030.log
Weights: https://download.openmmlab.com/mmaction/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics_rgb/omnisource/slowonly_r50_8x8x1_256e_minikinetics_omnisource_rgb_20201030-284cfd3b.pth
- Config: configs/recognition/tsn/tsn_r50_1x1x3_100e_kinetics400_rgb.py
In Collection: OmniSource
Metadata:
Architecture: ResNet50
Batch Size: 32
Epochs: 100
FLOPs: 102997721600
Parameters: 24327632
Pretrained: ImageNet
Resolution: 340x256
Training Data: Kinetics-400
Modality: RGB
Name: tsn_omnisource_r50_1x1x3_100e_kinetics_rgb
Converted From:
Weights: https://open-mmlab.s3.ap-northeast-2.amazonaws.com/mmaction/models/kinetics400/omnisource/tsn_OmniSource_kinetics400_se_rgb_r50_seg3_f1s1_imagenet-4066cb7e.pth
Code: https://github.com/open-mmlab/mmaction
Results:
- Dataset: Kinetics-400
Metrics:
Top 1 Accuracy: 73.6
Top 5 Accuracy: 91.0
Task: Action Recognition
Weights: https://download.openmmlab.com/mmaction/recognition/tsn/omni/tsn_imagenet_pretrained_r50_omni_1x1x3_kinetics400_rgb_20200926-54192355.pth
- Config: configs/recognition/tsn/tsn_r50_1x1x3_100e_kinetics400_rgb.py
In Collection: OmniSource
Metadata:
Architecture: ResNet50
Batch Size: 32
Epochs: 100
FLOPs: 102997721600
Parameters: 24327632
Pretrained: IG-1B
Resolution: short-side 320
Training Data: Kinetics-400
Modality: RGB
Name: tsn_IG1B_pretrained_omnisource_r50_1x1x3_100e_kinetics_rgb
Converted From:
Weights: https://open-mmlab.s3.ap-northeast-2.amazonaws.com/mmaction/models/kinetics400/omnisource/tsn_OmniSource_kinetics400_se_rgb_r50_seg3_f1s1_IG1B-25fc136b.pth
Code: https://github.com/open-mmlab/mmaction/
Results:
- Dataset: Kinetics-400
Metrics:
Top 1 Accuracy: 75.7
Top 5 Accuracy: 91.9
Task: Action Recognition
Weights: https://download.openmmlab.com/mmaction/recognition/tsn/omni/tsn_1G1B_pretrained_r50_omni_1x1x3_kinetics400_rgb_20200926-2863fed0.pth
- Config: configs/recognition/slowonly/slowonly_r50_4x16x1_256e_kinetics400_rgb.py
In Collection: OmniSource
Metadata:
Architecture: ResNet50
Batch Size: 8
Epochs: 256
FLOPs: 27430649856
Parameters: 32454096
Pretrained: None
Resolution: short-side 320
Training Data: Kinetics-400
Modality: RGB
Name: slowonly_r50_omnisource_4x16x1_256e_kinetics400_rgb
Converted From:
Weights: https://open-mmlab.s3.ap-northeast-2.amazonaws.com/mmaction/models/kinetics400/omnisource/slowonly_OmniSource_kinetics400_se_rgb_r50_seg1_4x16_scratch-71f7b8ee.pth
Code: https://github.com/open-mmlab/mmaction/
Results:
- Dataset: Kinetics-400
Metrics:
Top 1 Accuracy: 76.8
Top 5 Accuracy: 92.5
Task: Action Recognition
Weights: https://download.openmmlab.com/mmaction/recognition/slowonly/omni/slowonly_r50_omni_4x16x1_kinetics400_rgb_20200926-51b1f7ea.pth
- Config: configs/recognition/slowonly/slowonly_r101_8x8x1_196e_kinetics400_rgb.py
In Collection: OmniSource
Metadata:
Architecture: ResNet101
Batch Size: 8
Epochs: 196
FLOPs: 112063447040
Parameters: 60359120
Pretrained: None
Resolution: short-side 320
Training Data: Kinetics-400
Modality: RGB
Name: slowonly_r101_omnisource_8x8x1_196e_kinetics400_rgb
Converted From:
Weights: https://open-mmlab.s3.ap-northeast-2.amazonaws.com/mmaction/models/kinetics400/omnisource/slowonly_OmniSource_kinetics400_se_rgb_r101_seg1_8x8_scratch-2f838cb0.pth
Code: https://github.com/open-mmlab/mmaction/
Results:
- Dataset: Kinetics-400
Metrics:
Top 1 Accuracy: 80.4
Top 5 Accuracy: 94.4
Task: Action Recognition
Weights: https://download.openmmlab.com/mmaction/recognition/slowonly/omni/slowonly_r101_omni_8x8x1_kinetics400_rgb_20200926-b5dbb701.pth
configs/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics/slowonly_r50_8x8x1_256e_minikinetics_googleimage_rgb.py 0 → 100644
View file @ aa58d024
_base_ = [
'../../../_base_/models/slowonly_r50.py',
'../../../_base_/default_runtime.py'
]
# model settings
model = dict(backbone=dict(pretrained=None), cls_head=dict(num_classes=200))
# dataset settings
dataset_type = 'VideoDataset'
# The flag indicates using joint training
omnisource = True
data_root = 'data/OmniSource/kinetics_200_train'
data_root_val = 'data/OmniSource/kinetics_200_val'
gg_root = 'data/OmniSource/googleimage_200'
ann_file_train = 'data/OmniSource/annotations/kinetics_200/k200_train.txt'
ann_file_gg = ('data/OmniSource/annotations/googleimage_200/'
'tsn_8seg_googleimage_200_wodup.txt')
ann_file_val = 'data/OmniSource/annotations/kinetics_200/k200_val.txt'
ann_file_test = 'data/OmniSource/annotations/kinetics_200/k200_val.txt'
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_bgr=False)
train_pipeline = [
dict(type='DecordInit'),
dict(type='SampleFrames', clip_len=8, frame_interval=8, num_clips=1),
dict(type='DecordDecode'),
dict(type='Resize', scale=(-1, 256)),
dict(type='RandomResizedCrop'),
dict(type='Resize', scale=(224, 224), keep_ratio=False),
dict(type='Flip', flip_ratio=0.5),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs', 'label'])
]
train_gg_pipeline = [
dict(type='ImageDecode'),
dict(type='Resize', scale=(-1, 256)),
dict(type='RandomResizedCrop'),
dict(type='Resize', scale=(224, 224), keep_ratio=False),
dict(type='Flip', flip_ratio=0.5),
dict(type='BuildPseudoClip', clip_len=8),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs', 'label'])
]
val_pipeline = [
dict(type='DecordInit'),
dict(
type='SampleFrames',
clip_len=8,
frame_interval=8,
num_clips=1,
test_mode=True),
dict(type='DecordDecode'),
dict(type='Resize', scale=(-1, 256)),
dict(type='CenterCrop', crop_size=256),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs'])
]
test_pipeline = [
dict(type='DecordInit'),
dict(
type='SampleFrames',
clip_len=8,
frame_interval=8,
num_clips=10,
test_mode=True),
dict(type='DecordDecode'),
dict(type='Resize', scale=(-1, 256)),
dict(type='ThreeCrop', crop_size=256),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs'])
]
data = dict(
videos_per_gpu=12,
workers_per_gpu=2,
test_dataloader=dict(videos_per_gpu=1),
train=[
dict(
type=dataset_type,
ann_file=ann_file_train,
data_prefix=data_root,
pipeline=train_pipeline),
dict(
type='ImageDataset',
ann_file=ann_file_gg,
data_prefix=gg_root,
pipeline=train_gg_pipeline)
],
val=dict(
type=dataset_type,
ann_file=ann_file_val,
data_prefix=data_root_val,
pipeline=val_pipeline),
test=dict(
type=dataset_type,
ann_file=ann_file_test,
data_prefix=data_root_val,
pipeline=test_pipeline))
evaluation = dict(
interval=8, metrics=['top_k_accuracy', 'mean_class_accuracy'])
# optimizer
optimizer = dict(
type='SGD', lr=0.15, momentum=0.9,
weight_decay=0.0001) # this lr is used for 8 gpus
optimizer_config = dict(grad_clip=dict(max_norm=40, norm_type=2))
# learning policy
lr_config = dict(policy='CosineAnnealing', min_lr=0)
# runtime settings
total_epochs = 256
checkpoint_config = dict(interval=8)
work_dir = ('./work_dirs/omnisource/'
'slowonly_r50_8x8x1_256e_minikinetics_googleimage_rgb')
find_unused_parameters = False
configs/recognition/omnisource/slowonly_r50_8x8x1_256e_minikinetics/slowonly_r50_8x8x1_256e_minikinetics_insvideo_rgb.py 0 → 100644
View file @ aa58d024
_base_ = [
'../../../_base_/models/slowonly_r50.py',
'../../../_base_/default_runtime.py'
]
# model settings
model = dict(backbone=dict(pretrained=None), cls_head=dict(num_classes=200))
# dataset settings
dataset_type = 'VideoDataset'
# The flag indicates using joint training
omnisource = True
data_root = 'data/OmniSource/kinetics_200_train'
data_root_val = 'data/OmniSource/kinetics_200_val'
iv_root = 'data/OmniSource/insvideo_200'
ann_file_train = 'data/OmniSource/annotations/kinetics_200/k200_train.txt'
ann_file_iv = ('data/OmniSource/annotations/insvideo_200/'
'slowonly_8x8_insvideo_200_wodup.txt')
ann_file_val = 'data/OmniSource/annotations/kinetics_200/k200_val.txt'
ann_file_test = 'data/OmniSource/annotations/kinetics_200/k200_val.txt'
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_bgr=False)
train_pipeline = [
dict(type='DecordInit'),
dict(type='SampleFrames', clip_len=8, frame_interval=8, num_clips=1),
dict(type='DecordDecode'),
dict(type='Resize', scale=(-1, 256)),
dict(type='RandomResizedCrop'),
dict(type='Resize', scale=(224, 224), keep_ratio=False),
dict(type='Flip', flip_ratio=0.5),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs', 'label'])
]
train_iv_pipeline = [
dict(type='DecordInit'),
dict(type='SampleFrames', clip_len=8, frame_interval=8, num_clips=1),
dict(type='DecordDecode'),
dict(type='Resize', scale=(-1, 256)),
dict(type='RandomResizedCrop'),
dict(type='Resize', scale=(224, 224), keep_ratio=False),
dict(type='Flip', flip_ratio=0.5),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs', 'label'])
]
val_pipeline = [
dict(type='DecordInit'),
dict(
type='SampleFrames',
clip_len=8,
frame_interval=8,
num_clips=1,
test_mode=True),
dict(type='DecordDecode'),
dict(type='Resize', scale=(-1, 256)),
dict(type='CenterCrop', crop_size=256),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs'])
]
test_pipeline = [
dict(type='DecordInit'),
dict(
type='SampleFrames',
clip_len=8,
frame_interval=8,
num_clips=10,
test_mode=True),
dict(type='DecordDecode'),
dict(type='Resize', scale=(-1, 256)),
dict(type='ThreeCrop', crop_size=256),
dict(type='Normalize', **img_norm_cfg),
dict(type='FormatShape', input_format='NCTHW'),
dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
dict(type='ToTensor', keys=['imgs'])
]
data = dict(
videos_per_gpu=12,
workers_per_gpu=2,
test_dataloader=dict(videos_per_gpu=1),
train=[
dict(
type=dataset_type,
ann_file=ann_file_train,
data_prefix=data_root,
pipeline=train_pipeline),
dict(
type=dataset_type,
ann_file=ann_file_iv,
data_prefix=iv_root,
pipeline=train_iv_pipeline,
num_classes=200,
sample_by_class=True,
power=0.5)
],
val=dict(
type=dataset_type,
ann_file=ann_file_val,
data_prefix=data_root_val,
pipeline=val_pipeline),
test=dict(
type=dataset_type,
ann_file=ann_file_test,
data_prefix=data_root_val,
pipeline=test_pipeline))
evaluation = dict(
interval=8, metrics=['top_k_accuracy', 'mean_class_accuracy'])
# optimizer
optimizer = dict(
type='SGD', lr=0.15, momentum=0.9,
weight_decay=0.0001) # this lr is used for 8 gpus
optimizer_config = dict(grad_clip=dict(max_norm=40, norm_type=2))
# learning policy
lr_config = dict(policy='CosineAnnealing', min_lr=0)
# runtime settings
total_epochs = 256
checkpoint_config = dict(interval=8)
work_dir = ('./work_dirs/omnisource/'
'slowonly_r50_8x8x1_256e_minikinetics_insvideo_rgb')
find_unused_parameters = False
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment