Skip to content

GitLab

Commit 522a602f authored by wangkx1's avatar wangkx1
Browse files

siton bug

parent abb99c90
Hide whitespace changes
Inline Side-by-side
Showing with 748 additions and 0 deletions
configs/yolov3/yolov3_darknet53_270e_voc.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'../datasets/voc.yml',
'../runtime.yml',
'_base_/optimizer_270e.yml',
'_base_/yolov3_darknet53.yml',
'_base_/yolov3_reader.yml',
]
snapshot_epoch: 5
weights: output/yolov3_darknet53_270e_voc/model_final
pretrain_weights: https://paddledet.bj.bcebos.com/models/yolov3_darknet53_270e_coco.pdparams
# set collate_batch to false because ground-truth info is needed
# on voc dataset and should not collate data in batch when batch size
# is larger than 1.
EvalReader:
collate_batch: false
# ### remove comment below and run evaluate again to get 56.1 COCO for mAP(0.5:0.95)
# metric: COCO
# EvalDataset:
# !COCODataSet
# image_dir: VOCdevkit/VOC2007/JPEGImages
# anno_path: voc_test.json
# dataset_dir: dataset/voc
# # wget https://bj.bcebos.com/v1/paddledet/data/voc.zip
configs/yolov3/yolov3_darknet53_original_270e_coco.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_270e.yml',
'_base_/yolov3_darknet53.yml',
'_base_/yolov3_reader.yml',
]
snapshot_epoch: 5
weights: output/yolov3_darknet53_270e_coco/model_final
norm_type: bn
YOLOv3Loss:
ignore_thresh: 0.5
downsample: [32, 16, 8]
label_smooth: false
TrainReader:
inputs_def:
num_max_boxes: 50
sample_transforms:
- Decode: {}
- RandomDistort: {}
- RandomExpand: {fill_value: [123.675, 116.28, 103.53], ratio: 2.0}
- RandomCrop: {}
- RandomFlip: {}
batch_transforms:
- BatchRandomResize: {target_size: [320, 352, 384, 416, 448, 480, 512, 544, 576, 608], random_size: True, random_interp: True, keep_ratio: False}
- NormalizeBox: {}
- PadBox: {num_max_boxes: 50}
- BboxXYXY2XYWH: {}
- NormalizeImage: {mean: [0.485, 0.456, 0.406], std: [0.229, 0.224, 0.225], is_scale: True}
- Permute: {}
- Gt2YoloTarget: {anchor_masks: [[6, 7, 8], [3, 4, 5], [0, 1, 2]], anchors: [[10, 13], [16, 30], [33, 23], [30, 61], [62, 45], [59, 119], [116, 90], [156, 198], [373, 326]], downsample_ratios: [32, 16, 8], iou_thresh: 0.5}
batch_size: 8
shuffle: true
drop_last: true
mixup_epoch: -1
use_shared_memory: true
configs/yolov3/yolov3_darknet53_original_320e_coco_1p.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/yolov3_darknet53.yml',
'_base_/yolov3_reader.yml',
]
snapshot_epoch: 5
weights: output/yolov3_darknet53_270e_coco/model_final
norm_type: bn
YOLOv3Loss:
ignore_thresh: 0.5
downsample: [32, 16, 8]
label_smooth: false
worker_num: 8
TrainReader:
inputs_def:
num_max_boxes: 50
sample_transforms:
- Decode: {}
- RandomDistort: {}
- RandomExpand: {fill_value: [123.675, 116.28, 103.53], ratio: 2.0}
- RandomCrop: {}
- RandomFlip: {}
batch_transforms:
- BatchRandomResize: {target_size: [416], random_size: True, random_interp: True, keep_ratio: False}
- NormalizeBox: {}
- PadBox: {num_max_boxes: 50}
- BboxXYXY2XYWH: {}
- NormalizeImage: {mean: [0.485, 0.456, 0.406], std: [0.229, 0.224, 0.225], is_scale: True}
- Permute: {}
- Gt2YoloTarget: {anchor_masks: [[6, 7, 8], [3, 4, 5], [0, 1, 2]], anchors: [[10, 13], [16, 30], [33, 23], [30, 61], [62, 45], [59, 119], [116, 90], [156, 198], [373, 326]], downsample_ratios: [32, 16, 8], iou_thresh: 0.5}
batch_size: 32
shuffle: true
drop_last: true
mixup_epoch: -1
use_shared_memory: true
epoch: 320
LearningRate:
base_lr: 0.001
schedulers:
- !CosineDecay
max_epochs: 320
- !LinearWarmup
start_factor: 0.
epochs: 4
OptimizerBuilder:
optimizer:
momentum: 0.9
type: Momentum
regularizer:
factor: 0.016
type: L2
configs/yolov3/yolov3_mobilenet_v1_270e_coco.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_270e.yml',
'_base_/yolov3_mobilenet_v1.yml',
'_base_/yolov3_reader.yml',
]
snapshot_epoch: 5
weights: output/yolov3_mobilenet_v1_270e_coco/model_final
configs/yolov3/yolov3_mobilenet_v1_270e_voc.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'../datasets/voc.yml',
'../runtime.yml',
'_base_/optimizer_270e.yml',
'_base_/yolov3_mobilenet_v1.yml',
'_base_/yolov3_reader.yml',
]
snapshot_epoch: 5
weights: output/yolov3_mobilenet_v1_270e_voc/model_final
# set collate_batch to false because ground-truth info is needed
# on voc dataset and should not collate data in batch when batch size
# is larger than 1.
EvalReader:
collate_batch: false
LearningRate:
base_lr: 0.001
schedulers:
- !PiecewiseDecay
gamma: 0.1
milestones:
- 216
- 243
- !LinearWarmup
start_factor: 0.
steps: 1000
configs/yolov3/yolov3_mobilenet_v1_roadsign.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'../datasets/roadsign_voc.yml',
'../runtime.yml',
'_base_/optimizer_40e.yml',
'_base_/yolov3_mobilenet_v1.yml',
'_base_/yolov3_reader.yml',
]
pretrain_weights: https://paddledet.bj.bcebos.com/models/yolov3_mobilenet_v1_270e_coco.pdparams
weights: output/yolov3_mobilenet_v1_roadsign/model_final
YOLOv3Loss:
ignore_thresh: 0.7
label_smooth: true
configs/yolov3/yolov3_mobilenet_v1_ssld_270e_coco.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_270e.yml',
'_base_/yolov3_mobilenet_v1.yml',
'_base_/yolov3_reader.yml',
]
snapshot_epoch: 5
pretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/MobileNetV1_ssld_pretrained.pdparams
weights: output/yolov3_mobilenet_v1_ssld_270e_coco/model_final
configs/yolov3/yolov3_mobilenet_v1_ssld_270e_voc.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'../datasets/voc.yml',
'../runtime.yml',
'_base_/optimizer_270e.yml',
'_base_/yolov3_mobilenet_v1.yml',
'_base_/yolov3_reader.yml',
]
snapshot_epoch: 5
pretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/MobileNetV1_ssld_pretrained.pdparams
weights: output/yolov3_mobilenet_v1_ssld_270e_voc/model_final
# set collate_batch to false because ground-truth info is needed
# on voc dataset and should not collate data in batch when batch size
# is larger than 1.
EvalReader:
collate_batch: false
LearningRate:
base_lr: 0.001
schedulers:
- !PiecewiseDecay
gamma: 0.1
milestones:
- 216
- 243
- !LinearWarmup
start_factor: 0.
steps: 1000
configs/yolov3/yolov3_mobilenet_v3_large_270e_coco.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_270e.yml',
'_base_/yolov3_mobilenet_v3_large.yml',
'_base_/yolov3_reader.yml',
]
snapshot_epoch: 5
weights: output/yolov3_mobilenet_v3_large_270e_coco/model_final
configs/yolov3/yolov3_mobilenet_v3_large_270e_voc.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'../datasets/voc.yml',
'../runtime.yml',
'_base_/optimizer_270e.yml',
'_base_/yolov3_mobilenet_v3_large.yml',
'_base_/yolov3_reader.yml',
]
snapshot_epoch: 5
weights: output/yolov3_mobilenet_v3_large_270e_voc/model_final
# set collate_batch to false because ground-truth info is needed
# on voc dataset and should not collate data in batch when batch size
# is larger than 1.
EvalReader:
collate_batch: false
LearningRate:
base_lr: 0.001
schedulers:
- !PiecewiseDecay
gamma: 0.1
milestones:
- 216
- 243
- !LinearWarmup
start_factor: 0.
steps: 1000
configs/yolov3/yolov3_mobilenet_v3_large_ssld_270e_voc.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'../datasets/voc.yml',
'../runtime.yml',
'_base_/optimizer_270e.yml',
'_base_/yolov3_mobilenet_v3_large.yml',
'_base_/yolov3_reader.yml',
]
snapshot_epoch: 5
pretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/MobileNetV3_large_x1_0_ssld_pretrained.pdparams
weights: output/yolov3_mobilenet_v3_large_ssld_270e_voc/model_final
# set collate_batch to false because ground-truth info is needed
# on voc dataset and should not collate data in batch when batch size
# is larger than 1.
EvalReader:
collate_batch: false
LearningRate:
base_lr: 0.001
schedulers:
- !PiecewiseDecay
gamma: 0.1
milestones:
- 216
- 243
- !LinearWarmup
start_factor: 0.
steps: 1000
configs/yolov3/yolov3_r34_270e_coco.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_270e.yml',
'_base_/yolov3_r34.yml',
'_base_/yolov3_reader.yml',
]
snapshot_epoch: 5
weights: output/yolov3_r34_270e_coco/model_final
configs/yolov3/yolov3_r50vd_dcn_270e_coco.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_270e.yml',
'_base_/yolov3_r50vd_dcn.yml',
'_base_/yolov3_reader.yml',
]
snapshot_epoch: 5
weights: output/yolov3_r50vd_dcn_270e_coco/model_final
configs/yolov5/README.md 0 → 100644
View file @ 522a602f
# YOLOv5
## 内容
- [模型库](#模型库)
- [使用说明](#使用说明)
- [速度测试](#速度测试)
## 模型库
### 基础模型
| 网络网络 | 输入尺寸 | 图片数/GPU | 学习率策略 | 模型推理耗时(ms) | mAP<sup>val<br>0.5:0.95 | mAP<sup>val<br>0.5 | Params(M) | FLOPs(G) | 下载链接 | 配置文件 |
| :------------- | :------- | :-------: | :------: | :------------: | :---------------------: | :----------------: |:---------: | :------: |:---------------: |:-----: |
| YOLOv5-n | 640 | 16 | 300e | 1.5 | 28.0 | 45.7 | 1.87 | 4.52 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5_n_300e_coco.pdparams) | [配置文件](./yolov5_n_300e_coco.yml) |
| YOLOv5-s | 640 | 16 | 300e | 2.6 | 37.6 | 56.7 | 7.24 | 16.54 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5_s_300e_coco.pdparams) | [配置文件](./yolov5_s_300e_coco.yml) |
| YOLOv5-m | 640 | 16 | 300e | 5.2 | 45.4 | 64.1 | 21.19 | 49.08 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5_m_300e_coco.pdparams) | [配置文件](./yolov5_m_300e_coco.yml) |
| YOLOv5-l | 640 | 16 | 300e | 7.9 | 48.9 | 67.1 | 46.56 | 109.32 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5_l_300e_coco.pdparams) | [配置文件](./yolov5_l_300e_coco.yml) |
| YOLOv5-x | 640 | 16 | 300e | 13.7 | 50.6 | 68.7 | 86.75 | 205.92 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5_x_300e_coco.pdparams) | [配置文件](./yolov5_x_300e_coco.yml) |
| YOLOv5-s ConvNeXt| 640 | 8 | 36e | - | 42.4 | 65.3 | 34.54 | 17.96 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5_convnext_s_36e_coco.pdparams) | [配置文件](../convnext/yolov5_convnext_s_36e_coco.yml) |
### SSOD预训练模型
| 网络网络 | 输入尺寸 | 图片数/GPU | 学习率策略 | 模型推理耗时(ms) | mAP<sup>val<br>0.5:0.95 | mAP<sup>val<br>0.5 | Params(M) | FLOPs(G) | 下载链接 | 配置文件 |
| :------------- | :------- | :-------: | :------: | :------------: | :---------------------: | :----------------: |:---------: | :------: |:---------------: |:-----: |
| YOLOv5-s | 640 | 16 | 80e | 2.6 | 38.8 | 58.2 | 7.24 | 16.54 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5_s_80e_ssod_finetune_coco.pdparams) | [配置文件](./yolov5_s_80e_ssod_finetune_coco.yml) |
### Objects-365预训练模型
| 网络网络 | 输入尺寸 | 图片数/GPU | 学习率策略 | 模型推理耗时(ms) | mAP<sup>val<br>0.5:0.95 | mAP<sup>val<br>0.5 | Params(M) | FLOPs(G) | 下载链接 | 配置文件 |
| :------------- | :------- | :-------: | :------: | :------------: | :---------------------: | :----------------: |:---------: | :------: |:---------------: |:-----: |
| YOLOv5-l | 640 | 16 | 30e | 7.9 | 49.3 | 67.8 | 46.56 | 109.32 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5_l_30e_obj365_finetune_coco.pdparams) | [配置文件](./yolov5_l_30e_obj365_finetune_coco.yml) |
### P6大尺度模型
| 网络网络 | 输入尺寸 | 图片数/GPU | 学习率策略 | 模型推理耗时(ms) | mAP<sup>val<br>0.5:0.95 | mAP<sup>val<br>0.5 | Params(M) | FLOPs(G) | 下载链接 | 配置文件 |
| :------------- | :------- | :-------: | :------: | :------------: | :---------------------: | :----------------: |:---------: | :------: |:---------------: |:-----: |
| YOLOv5p6-n | 1280 | 16 | 300e | - | 35.9 | 54.2 | 3.25 | 9.23 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5p6_n_300e_coco.pdparams) | [配置文件](./yolov5p6_n_300e_coco.yml) |
| YOLOv5p6-s | 1280 | 16 | 300e | - | 44.5 | 63.3 | 12.63 | 33.81 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5p6_s_300e_coco.pdparams) | [配置文件](./yolov5p6_s_300e_coco.yml) |
| YOLOv5p6-m | 1280 | 16 | 300e | - | 51.1 | 69.0 | 35.73 | 100.21 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5p6_m_300e_coco.pdparams) | [配置文件](./yolov5p6_m_300e_coco.yml) |
| YOLOv5p6-l | 1280 | 8 | 300e | - | 53.4 | 71.0 | 76.77 | 223.09 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5p6_l_300e_coco.pdparams) | [配置文件](./yolov5p6_l_300e_coco.yml) |
| YOLOv5p6-x | 1280 | 8 | 300e | - | 54.7 | 72.4 | 140.80 | 420.03 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5p6_x_300e_coco.pdparams) | [配置文件](./yolov5p6_x_300e_coco.yml) |
### [YOLOv5u](../yolov5u)
| 网络网络 | 输入尺寸 | 图片数/GPU | 学习率策略 | 模型推理耗时(ms) | mAP<sup>val<br>0.5:0.95 | mAP<sup>val<br>0.5 | Params(M) | FLOPs(G) | 下载链接 | 配置文件 |
| :------------- | :------- | :-------: | :------: | :------------: | :---------------------: | :----------------: |:---------: | :------: |:---------------: |:-----: |
| YOLOv5u-n | 640 | 16 | 300e | 1.61 | 34.5 | 49.7 | 2.65 | 7.79 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5u_n_300e_coco.pdparams) | [配置文件](./yolov5u/yolov5u_n_300e_coco.yml) |
| YOLOv5u-s | 640 | 16 | 300e | 2.66 | 43.0 | 59.7 | 9.15 | 24.12 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5u_s_300e_coco.pdparams) | [配置文件](./yolov5u/yolov5u_s_300e_coco.yml) |
| YOLOv5u-m | 640 | 16 | 300e | 5.50 | 49.0 | 65.7 | 25.11 | 64.42 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5u_m_300e_coco.pdparams) | [配置文件](./yolov5u/yolov5u_m_300e_coco.yml) |
| YOLOv5u-l | 640 | 16 | 300e | 8.73 | 52.2 | 69.0 | 53.23 | 135.34 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5u_l_300e_coco.pdparams) | [配置文件](./yolov5u/yolov5u_l_300e_coco.yml) |
| YOLOv5u-x | 640 | 16 | 300e | 15.49 | 53.1 | 69.9 | 97.28 | 246.89 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5u_x_300e_coco.pdparams) | [配置文件](./yolov5u/yolov5u_x_300e_coco.yml) |
**注意:**
- YOLOv5模型训练使用COCO train2017作为训练集，Box AP为在COCO val2017上的`mAP(IoU=0.5:0.95)`结果；
- YOLOv5u 模型表示YOLOv5结构使用YOLOv8的head和loss，是Anchor Free的检测方案，具体可参照[YOLOv5u](../yolov5u)；
- YOLOv5模型训练过程中默认使用8 GPUs进行混合精度训练，默认lr为0.01为8卡总batch_size的设置，如果**GPU卡数**或者每卡**batch size**发生改动，也不需要改动学习率，但为了保证高精度最好使用**总batch size大于64**的配置去训练；
- 模型推理耗时(ms)为TensorRT-FP16下测试的耗时，不包含数据预处理和模型输出后处理(NMS)的耗时。测试采用单卡Tesla T4 GPU，batch size=1，测试环境为**paddlepaddle-2.3.2**, **CUDA 11.2**, **CUDNN 8.2**, **GCC-8.2**, **TensorRT 8.0.3.4**，具体请参考[速度测试](#速度测试)。
- 如果你设置了`--run_benchmark=True`, 你首先需要安装以下依赖`pip install pynvml psutil GPUtil`。
### 部署模型
| 网络模型 | 输入尺寸 | 导出后的权重(w/o NMS) | ONNX(w/o NMS) |
| :-------- | :--------: | :---------------------: | :----------------: |
| YOLOv5-n | 640 | [( w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_n_300e_coco_w_nms.zip) &#124; [( w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_n_300e_coco_wo_nms.zip) | [( w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_n_300e_coco_w_nms.onnx) &#124; [( w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_n_300e_coco_wo_nms.onnx) |
| YOLOv5-s | 640 | [( w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_s_300e_coco_w_nms.zip) &#124; [( w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_s_300e_coco_wo_nms.zip) | [( w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_s_300e_coco_w_nms.onnx) &#124; [( w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_s_300e_coco_wo_nms.onnx) |
| YOLOv5-m | 640 | [( w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_m_300e_coco_w_nms.zip) &#124; [( w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_m_300e_coco_wo_nms.zip) | [( w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_m_300e_coco_w_nms.onnx) &#124; [( w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_m_300e_coco_wo_nms.onnx) |
| YOLOv5-l | 640 | [( w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_l_300e_coco_w_nms.zip) &#124; [( w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_l_300e_coco_wo_nms.zip) | [( w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_l_300e_coco_w_nms.onnx) &#124; [( w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_l_300e_coco_wo_nms.onnx) |
| YOLOv5-x | 640 | [( w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_x_300e_coco_w_nms.zip) &#124; [( w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_x_300e_coco_wo_nms.zip) | [( w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_x_300e_coco_w_nms.onnx) &#124; [( w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_x_300e_coco_wo_nms.onnx) |
## 使用教程
### 0. **一键运行全流程**
将以下命令写在一个脚本文件里如```run.sh```，一键运行命令为：```sh run.sh```，也可命令行一句句去运行。
```bash
model_name=yolov5 # 可修改，如 ppyoloe
job_name=yolov5_s_300e_coco # 可修改，如 ppyoloe_plus_crn_s_80e_coco
config=configs/${model_name}/${job_name}.yml
log_dir=log_dir/${job_name}
# weights=https://bj.bcebos.com/v1/paddledet/models/${job_name}.pdparams
weights=output/${job_name}/model_final.pdparams
# 1.训练（单卡/多卡），加 --eval 表示边训边评估，加 --amp 表示混合精度训练
# CUDA_VISIBLE_DEVICES=0 python tools/train.py -c ${config} --eval --amp
python -m paddle.distributed.launch --log_dir=${log_dir} --gpus 0,1,2,3,4,5,6,7 tools/train.py -c ${config} --eval --amp
# 2.评估，加 --classwise 表示输出每一类mAP
CUDA_VISIBLE_DEVICES=0 python tools/eval.py -c ${config} -o weights=${weights} --classwise
# 3.预测 (单张图/图片文件夹）
CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c ${config} -o weights=${weights} --infer_img=demo/000000014439_640x640.jpg --draw_threshold=0.5
# CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c ${config} -o weights=${weights} --infer_dir=demo/ --draw_threshold=0.5
# 4.导出模型，以下3种模式选一种
## 普通导出，加trt表示用于trt加速，对NMS和silu激活函数提速明显
CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c ${config} -o weights=${weights} # trt=True
## exclude_post_process去除后处理导出，返回和YOLOv5导出ONNX时相同格式的concat后的1个Tensor，是未缩放回原图的坐标+分类置信度
# CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c ${config} -o weights=${weights} exclude_post_process=True # trt=True
## exclude_nms去除NMS导出，返回2个Tensor，是缩放回原图后的坐标和分类置信度
# CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c ${config} -o weights=${weights} exclude_nms=True # trt=True
# 5.部署预测，注意不能使用 去除后处理 或 去除NMS 导出后的模型去预测
CUDA_VISIBLE_DEVICES=0 python deploy/python/infer.py --model_dir=output_inference/${job_name} --image_file=demo/000000014439_640x640.jpg --device=GPU
# 6.部署测速，加 “--run_mode=trt_fp16” 表示在TensorRT FP16模式下测速，注意如需用到 trt_fp16 则必须为加 trt=True 导出的模型
CUDA_VISIBLE_DEVICES=0 python deploy/python/infer.py --model_dir=output_inference/${job_name} --image_file=demo/000000014439_640x640.jpg --device=GPU --run_benchmark=True # --run_mode=trt_fp16
# 7.onnx导出，一般结合 exclude_post_process去除后处理导出的模型
paddle2onnx --model_dir output_inference/${job_name} --model_filename model.pdmodel --params_filename model.pdiparams --opset_version 12 --save_file ${job_name}.onnx
# 8.onnx trt测速
/usr/local/TensorRT-8.0.3.4/bin/trtexec --onnx=${job_name}.onnx --workspace=4096 --avgRuns=10 --shapes=input:1x3x640x640 --fp16
/usr/local/TensorRT-8.0.3.4/bin/trtexec --onnx=${job_name}.onnx --workspace=4096 --avgRuns=10 --shapes=input:1x3x640x640 --fp32
```
### 1. 训练
执行以下指令使用混合精度训练YOLOv5
```bash
python -m paddle.distributed.launch --gpus 0,1,2,3,4,5,6,7 tools/train.py -c configs/yolov5/yolov5_s_300e_coco.yml --amp --eval
```
**注意:**
- `--amp`表示开启混合精度训练以避免显存溢出，`--eval`表示边训边验证。
### 2. 评估
执行以下命令在单个GPU上评估COCO val2017数据集
```bash
CUDA_VISIBLE_DEVICES=0 python tools/eval.py -c configs/yolov5/yolov5_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/yolov5_s_300e_coco.pdparams
```
### 3. 推理
使用以下命令在单张GPU上预测图片，使用`--infer_img`推理单张图片以及使用`--infer_dir`推理文件中的所有图片。
```bash
# 推理单张图片
CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c configs/yolov5/yolov5_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/yolov5_s_300e_coco.pdparams --infer_img=demo/000000014439_640x640.jpg
# 推理文件中的所有图片
CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c configs/yolov5/yolov5_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/yolov5_s_300e_coco.pdparams --infer_dir=demo
```
### 4.导出模型
YOLOv5在GPU上推理部署或benchmark测速等需要通过`tools/export_model.py`导出模型。
当你**使用Paddle Inference但不使用TensorRT**时，运行以下的命令导出模型
```bash
python tools/export_model.py -c configs/yolov5/yolov5_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/yolov5_s_300e_coco.pdparams
```
当你**使用Paddle Inference且使用TensorRT**时，需要指定`-o trt=True`来导出模型。
```bash
python tools/export_model.py -c configs/yolov5/yolov5_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/yolov5_s_300e_coco.pdparams trt=True
```
如果你想将YOLOv5模型导出为**ONNX格式**，参考
[PaddleDetection模型导出为ONNX格式教程](../../deploy/EXPORT_ONNX_MODEL.md)，运行以下命令：
```bash
# 导出推理模型
python tools/export_model.py -c configs/yolov5/yolov5_s_300e_coco.yml --output_dir=output_inference -o weights=https://paddledet.bj.bcebos.com/models/yolov5_s_300e_coco.pdparams
# 安装paddle2onnx
pip install paddle2onnx
# 转换成onnx格式
paddle2onnx --model_dir output_inference/yolov5_s_300e_coco --model_filename model.pdmodel --params_filename model.pdiparams --opset_version 11 --save_file yolov5_s_300e_coco.onnx
```
**注意：** ONNX模型目前只支持batch_size=1
### 5.推理部署
YOLOv5可以使用以下方式进行部署：
- Paddle Inference [Python](../../deploy/python) & [C++](../../deploy/cpp)
- [Paddle-TensorRT](../../deploy/TENSOR_RT.md)
- [PaddleServing](https://github.com/PaddlePaddle/Serving)
- [PaddleSlim模型量化](../slim)
运行以下命令导出模型
```bash
python tools/export_model.py -c configs/yolov5/yolov5_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/yolov5_s_300e_coco.pdparams trt=True
```
**注意：**
- trt=True表示**使用Paddle Inference且使用TensorRT**进行测速，速度会更快，默认不加即为False，表示**使用Paddle Inference但不使用TensorRT**进行测速。
- 如果是使用Paddle Inference在TensorRT FP16模式下部署，需要参考[Paddle Inference文档](https://www.paddlepaddle.org.cn/inference/master/user_guides/download_lib.html#python)，下载并安装与你的CUDA, CUDNN和TensorRT相应的wheel包。
#### 5.1.Python部署
`deploy/python/infer.py`使用上述导出后的Paddle Inference模型用于推理和benchnark测速，如果设置了`--run_benchmark=True`, 首先需要安装以下依赖`pip install pynvml psutil GPUtil`。
```bash
# Python部署推理单张图片
python deploy/python/infer.py --model_dir=output_inference/yolov5_s_300e_coco --image_file=demo/000000014439_640x640.jpg --device=gpu
# 推理文件夹下的所有图片
python deploy/python/infer.py --model_dir=output_inference/yolov5_s_300e_coco --image_dir=demo/ --device=gpu
```
#### 5.2. C++部署
`deploy/cpp/build/main`使用上述导出后的Paddle Inference模型用于C++推理部署, 首先按照[docs](../../deploy/cpp/docs)编译安装环境。
```bash
# C++部署推理单张图片
./deploy/cpp/build/main --model_dir=output_inference/yolov5_s_300e_coco/ --image_file=demo/000000014439_640x640.jpg --run_mode=paddle --device=GPU --threshold=0.5 --output_dir=cpp_infer_output/yolov5_s_300e_coco
```
## 速度测试
为了公平起见，在[模型库](#模型库)中的速度测试结果均为不包含数据预处理和模型输出后处理(NMS)的数据(与[YOLOv4(AlexyAB)](https://github.com/AlexeyAB/darknet)测试方法一致)，需要在导出模型时指定`-o exclude_nms=True`。测速需设置`--run_benchmark=True`, 首先需要安装以下依赖`pip install pynvml psutil GPUtil`。
**使用Paddle Inference但不使用TensorRT**进行测速，执行以下命令：
```bash
# 导出模型
python tools/export_model.py -c configs/yolov5/yolov5_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/yolov5_s_300e_coco.pdparams exclude_nms=True
# 速度测试，使用run_benchmark=True
python deploy/python/infer.py --model_dir=output_inference/yolov5_s_300e_coco --image_file=demo/000000014439_640x640.jpg --run_mode=paddle --device=gpu --run_benchmark=True
```
**使用Paddle Inference且使用TensorRT**进行测速，执行以下命令：
```bash
# 导出模型，使用trt=True
python tools/export_model.py -c configs/yolov5/yolov5_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/yolov5_s_300e_coco.pdparams exclude_nms=True trt=True
# 速度测试，使用run_benchmark=True
python deploy/python/infer.py --model_dir=output_inference/yolov5_s_300e_coco --image_file=demo/000000014439_640x640.jpg --device=gpu --run_benchmark=True
# tensorRT-FP32测速
python deploy/python/infer.py --model_dir=output_inference/yolov5_s_300e_coco --image_file=demo/000000014439_640x640.jpg --device=gpu --run_benchmark=True --run_mode=trt_fp32
# tensorRT-FP16测速
python deploy/python/infer.py --model_dir=output_inference/yolov5_s_300e_coco --image_file=demo/000000014439_640x640.jpg --device=gpu --run_benchmark=True --run_mode=trt_fp16
```
**注意:**
- 导出模型时指定`-o exclude_nms=True`仅作为测速时用，这样导出的模型其推理部署预测的结果不是最终检出框的结果。
- [模型库](#模型库)中的速度测试结果为tensorRT-FP16测速后的最快速度，为不包含数据预处理和模型输出后处理(NMS)的耗时。
configs/yolov5/_base_/optimizer_300e.yml 0 → 100644
View file @ 522a602f
epoch: 300
LearningRate:
base_lr: 0.01
schedulers:
- !YOLOv5LRDecay
max_epochs: 300
min_lr_ratio: 0.01
- !ExpWarmup
epochs: 3
OptimizerBuilder:
optimizer:
type: Momentum
momentum: 0.937
use_nesterov: True
regularizer:
factor: 0.0005
type: L2
configs/yolov5/_base_/optimizer_300e_high.yml 0 → 100644
View file @ 522a602f
epoch: 300
LearningRate:
base_lr: 0.01
schedulers:
- !YOLOv5LRDecay
max_epochs: 300
min_lr_ratio: 0.1 #
- !ExpWarmup
epochs: 3
OptimizerBuilder:
optimizer:
type: Momentum
momentum: 0.937
use_nesterov: True
regularizer:
factor: 0.0005
type: L2
configs/yolov5/_base_/yolov5_cspdarknet.yml 0 → 100644
View file @ 522a602f
architecture: YOLOv5
norm_type: sync_bn
use_ema: True
ema_decay: 0.9999
ema_decay_type: "exponential"
act: silu
find_unused_parameters: True
depth_mult: 1.0
width_mult: 1.0
YOLOv5:
backbone: CSPDarkNet
neck: YOLOCSPPAN
yolo_head: YOLOv5Head
post_process: ~
CSPDarkNet:
arch: "P5"
return_idx: [2, 3, 4]
depthwise: false
YOLOCSPPAN:
depthwise: false
YOLOv5Head:
anchors: [[10, 13], [16, 30], [33, 23],
[30, 61], [62, 45], [59, 119],
[116, 90], [156, 198], [373, 326]]
anchor_masks: [[0, 1, 2], [3, 4, 5], [6, 7, 8]]
stride: [8, 16, 32]
loss: YOLOv5Loss
nms:
name: MultiClassNMS
nms_top_k: 3000
keep_top_k: 300
score_threshold: 0.001
nms_threshold: 0.65
YOLOv5Loss:
downsample_ratios: [8, 16, 32]
balance: [4.0, 1.0, 0.4]
box_weight: 0.05
obj_weight: 1.0
cls_weght: 0.5
configs/yolov5/_base_/yolov5_cspresnet.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'yolov5_cspdarknet.yml',
]
YOLOv5:
backbone: CSPResNet
neck: CustomCSPPAN
yolo_head: YOLOv5Head
post_process: ~
CSPResNet:
layers: [3, 6, 6, 3]
channels: [64, 128, 256, 512, 1024]
return_idx: [1, 2, 3]
use_large_stem: True
CustomCSPPAN:
out_channels: [768, 384, 192]
stage_num: 1
block_num: 3
act: 'swish'
spp: true
configs/yolov5/_base_/yolov5_reader.yml 0 → 100644
View file @ 522a602f
input_height: &input_height 640
input_width: &input_width 640
input_size: &input_size [*input_height, *input_width]
mosaic_epoch: &mosaic_epoch 290 # last 10 epochs close mosaic, totally 300 epochs as default
worker_num: 4
TrainReader:
sample_transforms:
- DecodeNormResize: {target_size: *input_size, mosaic: True}
- MosaicPerspective: {mosaic_prob: 1.0, target_size: *input_size}
- RandomHSV: {hgain: 0.015, sgain: 0.7, vgain: 0.4}
- RandomFlip: {}
- BboxXYXY2XYWH: {}
- NormalizeBox: {}
- NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none}
- RGBReverse: {} # bgr->rgb
- Permute: {}
batch_size: 8
shuffle: True
drop_last: False
use_shared_memory: False
collate_batch: False
mosaic_epoch: *mosaic_epoch
EvalReader:
sample_transforms:
- Decode: {}
- Resize: {target_size: *input_size, keep_ratio: True, interp: 1}
- Pad: {size: *input_size, fill_value: [114., 114., 114.]}
- NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none}
- Permute: {}
batch_size: 1
TestReader:
inputs_def:
image_shape: [3, 640, 640]
sample_transforms:
- Decode: {}
- Resize: {target_size: *input_size, keep_ratio: True, interp: 1}
- Pad: {size: *input_size, fill_value: [114., 114., 114.]}
- NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none}
- Permute: {}
batch_size: 1
fuse_normalize: False
configs/yolov5/_base_/yolov5_reader_high_aug.yml 0 → 100644
View file @ 522a602f
input_height: &input_height 640
input_width: &input_width 640
input_size: &input_size [*input_height, *input_width]
mosaic_epoch: &mosaic_epoch 290 # last 10 epochs close mosaic, totally 300 epochs as default
worker_num: 8
TrainReader:
sample_transforms:
- DecodeNormResize: {target_size: *input_size, mosaic: True}
- MosaicPerspective: {mosaic_prob: 1.0, target_size: *input_size, scale: 0.9, mixup_prob: 0.1, copy_paste_prob: 0.1}
- RandomHSV: {hgain: 0.015, sgain: 0.7, vgain: 0.4}
- RandomFlip: {}
- BboxXYXY2XYWH: {}
- NormalizeBox: {}
- NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none}
- RGBReverse: {} # bgr->rgb
- Permute: {}
batch_size: 8
shuffle: True
drop_last: False
use_shared_memory: True
collate_batch: False
mosaic_epoch: *mosaic_epoch
EvalReader:
sample_transforms:
- Decode: {}
- Resize: {target_size: *input_size, keep_ratio: True, interp: 1}
- Pad: {size: *input_size, fill_value: [114., 114., 114.]}
- NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none}
- Permute: {}
batch_size: 1
TestReader:
inputs_def:
image_shape: [3, 640, 640]
sample_transforms:
- Decode: {}
- Resize: {target_size: *input_size, keep_ratio: True, interp: 1}
- Pad: {size: *input_size, fill_value: [114., 114., 114.]}
- NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none}
- Permute: {}
batch_size: 1
fuse_normalize: 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