We implement RegNetX and RegNetY models in detection systems and provide their first results on Mask R-CNN, Faster R-CNN and RetinaNet.
The pre-trained modles are converted from [model zoo of pycls](https://github.com/facebookresearch/pycls/blob/master/MODEL_ZOO.md).
```latex
@article{radosavovic2020designing,
title={Designing Network Design Spaces},
author={Ilija Radosavovic and Raj Prateek Kosaraju and Ross Girshick and Kaiming He and Piotr Dollár},
year={2020},
eprint={2003.13678},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
```
## Usage
To use a regnet model, there are two steps to do:
1. Convert the model to ResNet-style supported by MMDetection
2. Modify backbone and neck in config accordingly
### Convert model
We already prepare models of FLOPs from 400M to 12G in our model zoo.
For more general usage, we also provide script `regnet2mmdet.py` in the tools directory to convert the key of models pretrained by [pycls](https://github.com/facebookresearch/pycls/) to
This script convert model from `PRETRAIN_PATH` and store the converted model in `STORE_PATH`.
### Modify config
The users can modify the config's `depth` of backbone and corresponding keys in `arch` according to the configs in the [pycls model zoo](https://github.com/facebookresearch/pycls/blob/master/MODEL_ZOO.md).
The parameter `in_channels` in FPN can be found in the Figure 15 & 16 of the paper (`wi` in the legend).
This directory already provides some configs with their performance, using RegNetX from 800MF to 12GF level.
For other pre-trained models or self-implemented regnet models, the users are responsible to check these parameters by themselves.
**Note**: Although Fig. 15 & 16 also provide `w0`, `wa`, `wm`, `group_w`, and `bot_mul` for `arch`, they are quantized thus inaccurate, using them sometimes produces different backbone that does not match the key in the pre-trained model.
## Results
### Mask R-CNN
| Backbone | Style | Lr schd | Mem (GB) | Inf time (fps) | box AP | mask AP | Config | Download |
1. The models are trained using a different weight decay, i.e., `weight_decay=5e-5` according to the setting in ImageNet training. This brings improvement of at least 0.7 AP absolute but does not improve the model using ResNet-50.
2. RetinaNets using RegNets are trained with learning rate 0.02 with gradient clip. We find that using learning rate 0.02 could improve the results by at least 0.7 AP absolute and gradient clip is necessary to stabilize the training. However, this does not improve the performance of ResNet-50-FPN RetinaNet.
