未验证 提交 a718694c 编写于 作者: W wangxinxin08 提交者: GitHub

add ppyolov2 (#2626)

* add ppyolov2

* fix bugs and modify docs

* modify code and doc according to review

* fix bugs while resolving conflicts
上级 28fd7bcb
...@@ -38,17 +38,20 @@ PP-YOLO improved performance and speed of YOLOv3 with following methods: ...@@ -38,17 +38,20 @@ PP-YOLO improved performance and speed of YOLOv3 with following methods:
| Model | GPU number | images/GPU | backbone | input shape | Box AP<sup>val</sup> | Box AP<sup>test</sup> | V100 FP32(FPS) | V100 TensorRT FP16(FPS) | download | config | | Model | GPU number | images/GPU | backbone | input shape | Box AP<sup>val</sup> | Box AP<sup>test</sup> | V100 FP32(FPS) | V100 TensorRT FP16(FPS) | download | config |
|:------------------------:|:-------:|:-------------:|:----------:| :-------:| :------------------: | :-------------------: | :------------: | :---------------------: | :------: | :------: | |:------------------------:|:-------:|:-------------:|:----------:| :-------:| :------------------: | :-------------------: | :------------: | :---------------------: | :------: | :------: |
| PP-YOLO | 8 | 24 | ResNet50vd | 608 | 44.8 | 45.2 | 72.9 | 155.6 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_1x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_r50vd_dcn_1x_coco.yml) | | PP-YOLO | 8 | 24 | ResNet50vd | 608 | 44.8 | 45.2 | 72.9 | 155.6 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_1x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_r50vd_dcn_1x_coco.yml) |
| PP-YOLO | 8 | 24 | ResNet50vd | 512 | 43.9 | 44.4 | 89.9 | 188.4 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_1x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_r50vd_dcn_1x_coco.yml) | | PP-YOLO | 8 | 24 | ResNet50vd | 512 | 43.9 | 44.4 | 89.9 | 188.4 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_1x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_r50vd_dcn_1x_coco.yml) |
| PP-YOLO | 8 | 24 | ResNet50vd | 416 | 42.1 | 42.5 | 109.1 | 215.4 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_1x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_r50vd_dcn_1x_coco.yml) | | PP-YOLO | 8 | 24 | ResNet50vd | 416 | 42.1 | 42.5 | 109.1 | 215.4 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_1x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_r50vd_dcn_1x_coco.yml) |
| PP-YOLO | 8 | 24 | ResNet50vd | 320 | 38.9 | 39.3 | 132.2 | 242.2 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_1x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_r50vd_dcn_1x_coco.yml) | | PP-YOLO | 8 | 24 | ResNet50vd | 320 | 38.9 | 39.3 | 132.2 | 242.2 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_1x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_r50vd_dcn_1x_coco.yml) |
| PP-YOLO_2x | 8 | 24 | ResNet50vd | 608 | 45.3 | 45.9 | 72.9 | 155.6 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_2x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_r50vd_dcn_2x_coco.yml) | | PP-YOLO_2x | 8 | 24 | ResNet50vd | 608 | 45.3 | 45.9 | 72.9 | 155.6 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_2x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_r50vd_dcn_2x_coco.yml) |
| PP-YOLO_2x | 8 | 24 | ResNet50vd | 512 | 44.4 | 45.0 | 89.9 | 188.4 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_2x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_r50vd_dcn_2x_coco.yml) | | PP-YOLO_2x | 8 | 24 | ResNet50vd | 512 | 44.4 | 45.0 | 89.9 | 188.4 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_2x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_r50vd_dcn_2x_coco.yml) |
| PP-YOLO_2x | 8 | 24 | ResNet50vd | 416 | 42.7 | 43.2 | 109.1 | 215.4 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_2x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_r50vd_dcn_2x_coco.yml) | | PP-YOLO_2x | 8 | 24 | ResNet50vd | 416 | 42.7 | 43.2 | 109.1 | 215.4 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_2x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_r50vd_dcn_2x_coco.yml) |
| PP-YOLO_2x | 8 | 24 | ResNet50vd | 320 | 39.5 | 40.1 | 132.2 | 242.2 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_2x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_r50vd_dcn_2x_coco.yml) | | PP-YOLO_2x | 8 | 24 | ResNet50vd | 320 | 39.5 | 40.1 | 132.2 | 242.2 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_2x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_r50vd_dcn_2x_coco.yml) |
| PP-YOLO_ResNet18vd | 4 | 32 | ResNet18vd | 512 | 29.2 | 29.5 | 357.1 | 657.9 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r18vd_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_r18vd_coco.yml) | | PP-YOLO | 4 | 32 | ResNet18vd | 512 | 29.2 | 29.5 | 357.1 | 657.9 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r18vd_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_r18vd_coco.yml) |
| PP-YOLO_ResNet18vd | 4 | 32 | ResNet18vd | 416 | 28.6 | 28.9 | 409.8 | 719.4 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r18vd_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_r18vd_coco.yml) | | PP-YOLO | 4 | 32 | ResNet18vd | 416 | 28.6 | 28.9 | 409.8 | 719.4 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r18vd_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_r18vd_coco.yml) |
| PP-YOLO_ResNet18vd | 4 | 32 | ResNet18vd | 320 | 26.2 | 26.4 | 480.7 | 763.4 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r18vd_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_r18vd_coco.yml) | | PP-YOLO | 4 | 32 | ResNet18vd | 320 | 26.2 | 26.4 | 480.7 | 763.4 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r18vd_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_r18vd_coco.yml) |
| PP-YOLOv2 | 8 | 12 | ResNet50vd | 640 | 49.1 | 49.5 | - | - | [model](https://paddledet.bj.bcebos.com/models/ppyolov2_r50vd_dcn_365e_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolov2_r50vd_dcn_365e_coco.yml) |
| PP-YOLOv2 | 8 | 12 | ResNet101vd | 640 | 49.7 | 50.1 | - | - | [model](https://paddledet.bj.bcebos.com/models/ppyolov2_r101vd_dcn_365e_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolov2_r101vd_dcn_365e_coco.yml) |
**Notes:** **Notes:**
...@@ -62,8 +65,8 @@ PP-YOLO improved performance and speed of YOLOv3 with following methods: ...@@ -62,8 +65,8 @@ PP-YOLO improved performance and speed of YOLOv3 with following methods:
| Model | GPU number | images/GPU | Model Size | input shape | Box AP<sup>val</sup> | Box AP50<sup>val</sup> | Kirin 990 1xCore(FPS) | download | config | | Model | GPU number | images/GPU | Model Size | input shape | Box AP<sup>val</sup> | Box AP50<sup>val</sup> | Kirin 990 1xCore(FPS) | download | config |
|:----------------------------:|:-------:|:-------------:|:----------:| :-------:| :------------------: | :--------------------: | :--------------------: | :------: | :------: | |:----------------------------:|:-------:|:-------------:|:----------:| :-------:| :------------------: | :--------------------: | :--------------------: | :------: | :------: |
| PP-YOLO_MobileNetV3_large | 4 | 32 | 28MB | 320 | 23.2 | 42.6 | 14.1 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_mbv3_large_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_mbv3_large_coco.yml) | | PP-YOLO_MobileNetV3_large | 4 | 32 | 28MB | 320 | 23.2 | 42.6 | 14.1 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_mbv3_large_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_mbv3_large_coco.yml) |
| PP-YOLO_MobileNetV3_small | 4 | 32 | 16MB | 320 | 17.2 | 33.8 | 21.5 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_mbv3_small_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_mbv3_small_coco.yml) | | PP-YOLO_MobileNetV3_small | 4 | 32 | 16MB | 320 | 17.2 | 33.8 | 21.5 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_mbv3_small_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_mbv3_small_coco.yml) |
**Notes:** **Notes:**
...@@ -91,9 +94,9 @@ PP-YOLO trained on Pascal VOC dataset as follows: ...@@ -91,9 +94,9 @@ PP-YOLO trained on Pascal VOC dataset as follows:
| Model | GPU number | images/GPU | backbone | input shape | Box AP50<sup>val</sup> | download | config | | Model | GPU number | images/GPU | backbone | input shape | Box AP50<sup>val</sup> | download | config |
|:------------------:|:----------:|:----------:|:----------:| :----------:| :--------------------: | :------: | :-----: | |:------------------:|:----------:|:----------:|:----------:| :----------:| :--------------------: | :------: | :-----: |
| PP-YOLO | 8 | 12 | ResNet50vd | 608 | 84.9 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_voc.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_r50vd_dcn_voc.yml) | | PP-YOLO | 8 | 12 | ResNet50vd | 608 | 84.9 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_voc.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_r50vd_dcn_voc.yml) |
| PP-YOLO | 8 | 12 | ResNet50vd | 416 | 84.3 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_voc.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_r50vd_dcn_voc.yml) | | PP-YOLO | 8 | 12 | ResNet50vd | 416 | 84.3 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_voc.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_r50vd_dcn_voc.yml) |
| PP-YOLO | 8 | 12 | ResNet50vd | 320 | 82.2 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_voc.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_r50vd_dcn_voc.yml) | | PP-YOLO | 8 | 12 | ResNet50vd | 320 | 82.2 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_voc.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_r50vd_dcn_voc.yml) |
## Getting Start ## Getting Start
...@@ -184,8 +187,7 @@ Optimizing method and ablation experiments of PP-YOLO compared with YOLOv3. ...@@ -184,8 +187,7 @@ Optimizing method and ablation experiments of PP-YOLO compared with YOLOv3.
- Performance and inference spedd are measure with input shape as 608 - Performance and inference spedd are measure with input shape as 608
- All models are trained on COCO train2017 datast and evaluated on val2017 & test-dev2017 dataset,`Box AP` is evaluation results as `mAP(IoU=0.5:0.95)`. - All models are trained on COCO train2017 datast and evaluated on val2017 & test-dev2017 dataset,`Box AP` is evaluation results as `mAP(IoU=0.5:0.95)`.
- Inference speed is tested on single Tesla V100 with batch size as 1 following test method and environment configuration in benchmark above. - Inference speed is tested on single Tesla V100 with batch size as 1 following test method and environment configuration in benchmark above.
- [YOLOv3-DarkNet53](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/yolov3/yolov3_darknet53_270e_coco.yml) with mAP as 39.0 is optimized YOLOv3 model in PaddleDetection,see [Model Zoo](https://github.com/PaddlePaddle/PaddleDetection/blob/master/docs/MODEL_ZOO.md) for details. - [YOLOv3-DarkNet53](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/yolov3/yolov3_darknet53_270e_coco.yml) with mAP as 39.0 is optimized YOLOv3 model in PaddleDetection,see [Model Zoo](https://github.com/PaddlePaddle/PaddleDetection/blob/master/docs/MODEL_ZOO.md) for details.
## Citation ## Citation
......
...@@ -38,17 +38,19 @@ PP-YOLO从如下方面优化和提升YOLOv3模型的精度和速度: ...@@ -38,17 +38,19 @@ PP-YOLO从如下方面优化和提升YOLOv3模型的精度和速度:
| 模型 | GPU个数 | 每GPU图片个数 | 骨干网络 | 输入尺寸 | Box AP<sup>val</sup> | Box AP<sup>test</sup> | V100 FP32(FPS) | V100 TensorRT FP16(FPS) | 模型下载 | 配置文件 | | 模型 | GPU个数 | 每GPU图片个数 | 骨干网络 | 输入尺寸 | Box AP<sup>val</sup> | Box AP<sup>test</sup> | V100 FP32(FPS) | V100 TensorRT FP16(FPS) | 模型下载 | 配置文件 |
|:------------------------:|:-------:|:-------------:|:----------:| :-------:| :------------------: | :-------------------: | :------------: | :---------------------: | :------: | :------: | |:------------------------:|:-------:|:-------------:|:----------:| :-------:| :------------------: | :-------------------: | :------------: | :---------------------: | :------: | :------: |
| PP-YOLO | 8 | 24 | ResNet50vd | 608 | 44.8 | 45.2 | 72.9 | 155.6 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_1x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_r50vd_dcn_1x_coco.yml) | | PP-YOLO | 8 | 24 | ResNet50vd | 608 | 44.8 | 45.2 | 72.9 | 155.6 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_1x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_r50vd_dcn_1x_coco.yml) |
| PP-YOLO | 8 | 24 | ResNet50vd | 512 | 43.9 | 44.4 | 89.9 | 188.4 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_1x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_r50vd_dcn_1x_coco.yml) | | PP-YOLO | 8 | 24 | ResNet50vd | 512 | 43.9 | 44.4 | 89.9 | 188.4 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_1x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_r50vd_dcn_1x_coco.yml) |
| PP-YOLO | 8 | 24 | ResNet50vd | 416 | 42.1 | 42.5 | 109.1 | 215.4 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_1x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_r50vd_dcn_1x_coco.yml) | | PP-YOLO | 8 | 24 | ResNet50vd | 416 | 42.1 | 42.5 | 109.1 | 215.4 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_1x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_r50vd_dcn_1x_coco.yml) |
| PP-YOLO | 8 | 24 | ResNet50vd | 320 | 38.9 | 39.3 | 132.2 | 242.2 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_1x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_r50vd_dcn_1x_coco.yml) | | PP-YOLO | 8 | 24 | ResNet50vd | 320 | 38.9 | 39.3 | 132.2 | 242.2 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_1x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_r50vd_dcn_1x_coco.yml) |
| PP-YOLO_2x | 8 | 24 | ResNet50vd | 608 | 45.3 | 45.9 | 72.9 | 155.6 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_2x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_r50vd_dcn_2x_coco.yml) | | PP-YOLO_2x | 8 | 24 | ResNet50vd | 608 | 45.3 | 45.9 | 72.9 | 155.6 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_2x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_r50vd_dcn_2x_coco.yml) |
| PP-YOLO_2x | 8 | 24 | ResNet50vd | 512 | 44.4 | 45.0 | 89.9 | 188.4 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_2x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_r50vd_dcn_2x_coco.yml) | | PP-YOLO_2x | 8 | 24 | ResNet50vd | 512 | 44.4 | 45.0 | 89.9 | 188.4 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_2x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_r50vd_dcn_2x_coco.yml) |
| PP-YOLO_2x | 8 | 24 | ResNet50vd | 416 | 42.7 | 43.2 | 109.1 | 215.4 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_2x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_r50vd_dcn_2x_coco.yml) | | PP-YOLO_2x | 8 | 24 | ResNet50vd | 416 | 42.7 | 43.2 | 109.1 | 215.4 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_2x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_r50vd_dcn_2x_coco.yml) |
| PP-YOLO_2x | 8 | 24 | ResNet50vd | 320 | 39.5 | 40.1 | 132.2 | 242.2 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_2x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_r50vd_dcn_2x_coco.yml) | | PP-YOLO_2x | 8 | 24 | ResNet50vd | 320 | 39.5 | 40.1 | 132.2 | 242.2 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_2x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_r50vd_dcn_2x_coco.yml) |
| PP-YOLO_ResNet18vd | 4 | 32 | ResNet18vd | 512 | 29.2 | 29.5 | 357.1 | 657.9 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r18vd_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_r18vd_coco.yml) | | PP-YOLO | 4 | 32 | ResNet18vd | 512 | 29.2 | 29.5 | 357.1 | 657.9 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r18vd_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_r18vd_coco.yml) |
| PP-YOLO_ResNet18vd | 4 | 32 | ResNet18vd | 416 | 28.6 | 28.9 | 409.8 | 719.4 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r18vd_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_r18vd_coco.yml) | | PP-YOLO | 4 | 32 | ResNet18vd | 416 | 28.6 | 28.9 | 409.8 | 719.4 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r18vd_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_r18vd_coco.yml) |
| PP-YOLO_ResNet18vd | 4 | 32 | ResNet18vd | 320 | 26.2 | 26.4 | 480.7 | 763.4 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r18vd_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_r18vd_coco.yml) | | PP-YOLO | 4 | 32 | ResNet18vd | 320 | 26.2 | 26.4 | 480.7 | 763.4 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r18vd_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_r18vd_coco.yml) |
| PP-YOLOv2 | 8 | 12 | ResNet50vd | 640 | 49.1 | 49.5 | - | - | [model](https://paddledet.bj.bcebos.com/models/ppyolov2_r50vd_dcn_365e_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolov2_r50vd_dcn_365e_coco.yml) |
| PP-YOLOv2 | 8 | 12 | ResNet101vd | 640 | 49.7 | 50.1 | - | - | [model](https://paddledet.bj.bcebos.com/models/ppyolov2_r101vd_dcn_365e_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolov2_r101vd_dcn_365e_coco.yml) |
**注意:** **注意:**
...@@ -63,8 +65,8 @@ PP-YOLO从如下方面优化和提升YOLOv3模型的精度和速度: ...@@ -63,8 +65,8 @@ PP-YOLO从如下方面优化和提升YOLOv3模型的精度和速度:
| 模型 | GPU个数 | 每GPU图片个数 | 模型体积 | 输入尺寸 | Box AP<sup>val</sup> | Box AP50<sup>val</sup> | Kirin 990 1xCore (FPS) | 模型下载 | 配置文件 | | 模型 | GPU个数 | 每GPU图片个数 | 模型体积 | 输入尺寸 | Box AP<sup>val</sup> | Box AP50<sup>val</sup> | Kirin 990 1xCore (FPS) | 模型下载 | 配置文件 |
|:----------------------------:|:-------:|:-------------:|:----------:| :-------:| :------------------: | :--------------------: | :--------------------: | :------: | :------: | |:----------------------------:|:-------:|:-------------:|:----------:| :-------:| :------------------: | :--------------------: | :--------------------: | :------: | :------: |
| PP-YOLO_MobileNetV3_large | 4 | 32 | 28MB | 320 | 23.2 | 42.6 | 14.1 | [下载链接](https://paddledet.bj.bcebos.com/models/ppyolo_mbv3_large_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_mbv3_large_coco.yml) | | PP-YOLO_MobileNetV3_large | 4 | 32 | 28MB | 320 | 23.2 | 42.6 | 14.1 | [下载链接](https://paddledet.bj.bcebos.com/models/ppyolo_mbv3_large_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_mbv3_large_coco.yml) |
| PP-YOLO_MobileNetV3_small | 4 | 32 | 16MB | 320 | 17.2 | 33.8 | 21.5 | [下载链接](https://paddledet.bj.bcebos.com/models/ppyolo_mbv3_small_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_mbv3_small_coco.yml) | | PP-YOLO_MobileNetV3_small | 4 | 32 | 16MB | 320 | 17.2 | 33.8 | 21.5 | [下载链接](https://paddledet.bj.bcebos.com/models/ppyolo_mbv3_small_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_mbv3_small_coco.yml) |
- PP-YOLO_MobileNetV3 模型使用COCO数据集中train2017作为训练集,使用val2017作为测试集,Box AP<sup>val</sup>`mAP(IoU=0.5:0.95)`评估结果, Box AP50<sup>val</sup>`mAP(IoU=0.5)`评估结果。 - PP-YOLO_MobileNetV3 模型使用COCO数据集中train2017作为训练集,使用val2017作为测试集,Box AP<sup>val</sup>`mAP(IoU=0.5:0.95)`评估结果, Box AP50<sup>val</sup>`mAP(IoU=0.5)`评估结果。
- PP-YOLO_MobileNetV3 模型训练过程中使用4GPU,每GPU batch size为32进行训练,如训练GPU数和batch size不使用上述配置,须参考[FAQ](https://github.com/PaddlePaddle/PaddleDetection/blob/master/docs/FAQ.md)调整学习率和迭代次数。 - PP-YOLO_MobileNetV3 模型训练过程中使用4GPU,每GPU batch size为32进行训练,如训练GPU数和batch size不使用上述配置,须参考[FAQ](https://github.com/PaddlePaddle/PaddleDetection/blob/master/docs/FAQ.md)调整学习率和迭代次数。
...@@ -76,9 +78,9 @@ PP-YOLO在Pascal VOC数据集上训练模型如下: ...@@ -76,9 +78,9 @@ PP-YOLO在Pascal VOC数据集上训练模型如下:
| 模型 | GPU个数 | 每GPU图片个数 | 骨干网络 | 输入尺寸 | Box AP50<sup>val</sup> | 模型下载 | 配置文件 | | 模型 | GPU个数 | 每GPU图片个数 | 骨干网络 | 输入尺寸 | Box AP50<sup>val</sup> | 模型下载 | 配置文件 |
|:------------------:|:-------:|:-------------:|:----------:| :----------:| :--------------------: | :------: | :-----: | |:------------------:|:-------:|:-------------:|:----------:| :----------:| :--------------------: | :------: | :-----: |
| PP-YOLO | 8 | 12 | ResNet50vd | 608 | 84.9 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_voc.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_r50vd_dcn_voc.yml) | | PP-YOLO | 8 | 12 | ResNet50vd | 608 | 84.9 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_voc.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_r50vd_dcn_voc.yml) |
| PP-YOLO | 8 | 12 | ResNet50vd | 416 | 84.3 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_voc.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_r50vd_dcn_voc.yml) | | PP-YOLO | 8 | 12 | ResNet50vd | 416 | 84.3 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_voc.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_r50vd_dcn_voc.yml) |
| PP-YOLO | 8 | 12 | ResNet50vd | 320 | 82.2 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_voc.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/ppyolo/ppyolo_r50vd_dcn_voc.yml) | | PP-YOLO | 8 | 12 | ResNet50vd | 320 | 82.2 | [model](https://paddledet.bj.bcebos.com/models/ppyolo_r50vd_dcn_voc.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppyolo/ppyolo_r50vd_dcn_voc.yml) |
## 使用说明 ## 使用说明
...@@ -169,8 +171,7 @@ PP-YOLO模型相对于YOLOv3模型优化项消融实验数据如下表所示。 ...@@ -169,8 +171,7 @@ PP-YOLO模型相对于YOLOv3模型优化项消融实验数据如下表所示。
- 精度与推理速度数据均为使用输入图像尺寸为608的测试结果 - 精度与推理速度数据均为使用输入图像尺寸为608的测试结果
- Box AP为在COCO train2017数据集训练,val2017和test-dev2017数据集上评估`mAP(IoU=0.5:0.95)`数据 - Box AP为在COCO train2017数据集训练,val2017和test-dev2017数据集上评估`mAP(IoU=0.5:0.95)`数据
- 推理速度为单卡V100上,batch size=1, 使用上述benchmark测试方法的测试结果,测试环境配置为CUDA 10.2,CUDNN 7.5.1 - 推理速度为单卡V100上,batch size=1, 使用上述benchmark测试方法的测试结果,测试环境配置为CUDA 10.2,CUDNN 7.5.1
- [YOLOv3-DarkNet53](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/yolov3/yolov3_darknet53_270e_coco.yml)精度38.9为PaddleDetection优化后的YOLOv3模型,可参见[模型库](https://github.com/PaddlePaddle/PaddleDetection/blob/master/docs/MODEL_ZOO.md) - [YOLOv3-DarkNet53](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/yolov3/yolov3_darknet53_270e_coco.yml)精度38.9为PaddleDetection优化后的YOLOv3模型,可参见[模型库](https://github.com/PaddlePaddle/PaddleDetection/blob/master/docs/MODEL_ZOO.md)
## 引用 ## 引用
......
epoch: 365
LearningRate:
base_lr: 0.005
schedulers:
- !PiecewiseDecay
gamma: 0.1
milestones:
- 243
- !LinearWarmup
start_factor: 0.
steps: 4000
OptimizerBuilder:
clip_grad_by_norm: 35.
optimizer:
momentum: 0.9
type: Momentum
regularizer:
factor: 0.0005
type: L2
architecture: YOLOv3
pretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/ResNet50_vd_ssld_pretrained.pdparams
norm_type: sync_bn
use_ema: true
ema_decay: 0.9998
YOLOv3:
backbone: ResNet
neck: PPYOLOPAN
yolo_head: YOLOv3Head
post_process: BBoxPostProcess
ResNet:
depth: 50
variant: d
return_idx: [1, 2, 3]
dcn_v2_stages: [3]
freeze_at: -1
freeze_norm: false
norm_decay: 0.
PPYOLOPAN:
drop_block: true
block_size: 3
keep_prob: 0.9
spp: true
YOLOv3Head:
anchors: [[10, 13], [16, 30], [33, 23],
[30, 61], [62, 45], [59, 119],
[116, 90], [156, 198], [373, 326]]
anchor_masks: [[6, 7, 8], [3, 4, 5], [0, 1, 2]]
loss: YOLOv3Loss
iou_aware: true
iou_aware_factor: 0.5
YOLOv3Loss:
ignore_thresh: 0.7
downsample: [32, 16, 8]
label_smooth: false
scale_x_y: 1.05
iou_loss: IouLoss
iou_aware_loss: IouAwareLoss
IouLoss:
loss_weight: 2.5
loss_square: true
IouAwareLoss:
loss_weight: 1.0
BBoxPostProcess:
decode:
name: YOLOBox
conf_thresh: 0.01
downsample_ratio: 32
clip_bbox: true
scale_x_y: 1.05
nms:
name: MatrixNMS
keep_top_k: 100
score_threshold: 0.01
post_threshold: 0.01
nms_top_k: -1
background_label: -1
worker_num: 8
TrainReader:
inputs_def:
num_max_boxes: 100
sample_transforms:
- Decode: {}
- Mixup: {alpha: 1.5, beta: 1.5}
- RandomDistort: {}
- RandomExpand: {fill_value: [123.675, 116.28, 103.53]}
- RandomCrop: {}
- RandomFlip: {}
batch_transforms:
- BatchRandomResize: {target_size: [320, 352, 384, 416, 448, 480, 512, 544, 576, 608, 640, 672, 704, 736, 768], random_size: True, random_interp: True, keep_ratio: False}
- NormalizeBox: {}
- PadBox: {num_max_boxes: 100}
- 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]}
batch_size: 12
shuffle: true
drop_last: true
mixup_epoch: 25000
use_shared_memory: true
EvalReader:
sample_transforms:
- Decode: {}
- Resize: {target_size: [640, 640], keep_ratio: False, interp: 2}
- NormalizeImage: {mean: [0.485, 0.456, 0.406], std: [0.229, 0.224, 0.225], is_scale: True}
- Permute: {}
batch_size: 8
drop_empty: false
TestReader:
inputs_def:
image_shape: [3, 640, 640]
sample_transforms:
- Decode: {}
- Resize: {target_size: [640, 640], keep_ratio: False, interp: 2}
- NormalizeImage: {mean: [0.485, 0.456, 0.406], std: [0.229, 0.224, 0.225], is_scale: True}
- Permute: {}
batch_size: 1
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'./_base_/ppyolov2_r50vd_dcn.yml',
'./_base_/optimizer_365e.yml',
'./_base_/ppyolov2_reader.yml',
]
snapshot_epoch: 8
weights: output/ppyolov2_r101vd_dcn_365e_coco/model_final
pretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/ResNet101_vd_ssld_pretrained.pdparams
ResNet:
depth: 101
variant: d
return_idx: [1, 2, 3]
dcn_v2_stages: [3]
freeze_at: -1
freeze_norm: false
norm_decay: 0.
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'./_base_/ppyolov2_r50vd_dcn.yml',
'./_base_/optimizer_365e.yml',
'./_base_/ppyolov2_reader.yml',
]
snapshot_epoch: 8
weights: output/ppyolov2_r50vd_dcn_365e_coco/model_final
...@@ -18,7 +18,7 @@ import paddle.nn.functional as F ...@@ -18,7 +18,7 @@ import paddle.nn.functional as F
from paddle import ParamAttr from paddle import ParamAttr
from paddle.regularizer import L2Decay from paddle.regularizer import L2Decay
from ppdet.core.workspace import register, serializable from ppdet.core.workspace import register, serializable
from ppdet.modeling.ops import batch_norm from ppdet.modeling.ops import batch_norm, mish
from ..shape_spec import ShapeSpec from ..shape_spec import ShapeSpec
__all__ = ['DarkNet', 'ConvBNLayer'] __all__ = ['DarkNet', 'ConvBNLayer']
...@@ -77,6 +77,8 @@ class ConvBNLayer(nn.Layer): ...@@ -77,6 +77,8 @@ class ConvBNLayer(nn.Layer):
out = self.batch_norm(out) out = self.batch_norm(out)
if self.act == 'leaky': if self.act == 'leaky':
out = F.leaky_relu(out, 0.1) out = F.leaky_relu(out, 0.1)
elif self.act == 'mish':
out = mish(out)
return out return out
......
...@@ -42,7 +42,7 @@ class IouAwareLoss(IouLoss): ...@@ -42,7 +42,7 @@ class IouAwareLoss(IouLoss):
iou = bbox_iou( iou = bbox_iou(
pbox, gbox, giou=self.giou, diou=self.diou, ciou=self.ciou) pbox, gbox, giou=self.giou, diou=self.diou, ciou=self.ciou)
iou.stop_gradient = True iou.stop_gradient = True
ioup = F.sigmoid(ioup) loss_iou_aware = F.binary_cross_entropy_with_logits(
loss_iou_aware = (-iou * paddle.log(ioup)).sum(-2, keepdim=True) ioup, iou, reduction='none')
loss_iou_aware = loss_iou_aware * self.loss_weight loss_iou_aware = loss_iou_aware * self.loss_weight
return loss_iou_aware return loss_iou_aware
...@@ -25,6 +25,32 @@ from ..shape_spec import ShapeSpec ...@@ -25,6 +25,32 @@ from ..shape_spec import ShapeSpec
__all__ = ['YOLOv3FPN', 'PPYOLOFPN'] __all__ = ['YOLOv3FPN', 'PPYOLOFPN']
def add_coord(x):
b = x.shape[0]
if self.data_format == 'NCHW':
h = x.shape[2]
w = x.shape[3]
else:
h = x.shape[1]
w = x.shape[2]
gx = paddle.arange(w, dtype='float32') / (w - 1.) * 2.0 - 1.
if self.data_format == 'NCHW':
gx = gx.reshape([1, 1, 1, w]).expand([b, 1, h, w])
else:
gx = gx.reshape([1, 1, w, 1]).expand([b, h, w, 1])
gx.stop_gradient = True
gy = paddle.arange(h, dtype='float32') / (h - 1.) * 2.0 - 1.
if self.data_format == 'NCHW':
gy = gy.reshape([1, 1, h, 1]).expand([b, 1, h, w])
else:
gy = gy.reshape([1, h, 1, 1]).expand([b, h, w, 1])
gy.stop_gradient = True
return gx, gy
class YoloDetBlock(nn.Layer): class YoloDetBlock(nn.Layer):
def __init__(self, ch_in, channel, norm_type, name, data_format='NCHW'): def __init__(self, ch_in, channel, norm_type, name, data_format='NCHW'):
""" """
...@@ -87,6 +113,7 @@ class SPP(nn.Layer): ...@@ -87,6 +113,7 @@ class SPP(nn.Layer):
pool_size, pool_size,
norm_type, norm_type,
name, name,
act='leaky',
data_format='NCHW'): data_format='NCHW'):
""" """
SPP layer, which consist of four pooling layer follwed by conv layer SPP layer, which consist of four pooling layer follwed by conv layer
...@@ -101,6 +128,7 @@ class SPP(nn.Layer): ...@@ -101,6 +128,7 @@ class SPP(nn.Layer):
""" """
super(SPP, self).__init__() super(SPP, self).__init__()
self.pool = [] self.pool = []
self.data_format = data_format
for size in pool_size: for size in pool_size:
pool = self.add_sublayer( pool = self.add_sublayer(
'{}.pool1'.format(name), '{}.pool1'.format(name),
...@@ -118,13 +146,18 @@ class SPP(nn.Layer): ...@@ -118,13 +146,18 @@ class SPP(nn.Layer):
padding=k // 2, padding=k // 2,
norm_type=norm_type, norm_type=norm_type,
name=name, name=name,
act=act,
data_format=data_format) data_format=data_format)
def forward(self, x): def forward(self, x):
outs = [x] outs = [x]
for pool in self.pool: for pool in self.pool:
outs.append(pool(x)) outs.append(pool(x))
y = paddle.concat(outs, axis=1) if self.data_format == "NCHW":
y = paddle.concat(outs, axis=1)
else:
y = paddle.concat(outs, axis=-1)
y = self.conv(y) y = self.conv(y)
return y return y
...@@ -204,28 +237,7 @@ class CoordConv(nn.Layer): ...@@ -204,28 +237,7 @@ class CoordConv(nn.Layer):
self.data_format = data_format self.data_format = data_format
def forward(self, x): def forward(self, x):
b = x.shape[0] gx, gy = add_coord(x)
if self.data_format == 'NCHW':
h = x.shape[2]
w = x.shape[3]
else:
h = x.shape[1]
w = x.shape[2]
gx = paddle.arange(w, dtype='float32') / (w - 1.) * 2.0 - 1.
if self.data_format == 'NCHW':
gx = gx.reshape([1, 1, 1, w]).expand([b, 1, h, w])
else:
gx = gx.reshape([1, 1, w, 1]).expand([b, h, w, 1])
gx.stop_gradient = True
gy = paddle.arange(h, dtype='float32') / (h - 1.) * 2.0 - 1.
if self.data_format == 'NCHW':
gy = gy.reshape([1, 1, h, 1]).expand([b, 1, h, w])
else:
gy = gy.reshape([1, h, 1, 1]).expand([b, h, w, 1])
gy.stop_gradient = True
if self.data_format == 'NCHW': if self.data_format == 'NCHW':
y = paddle.concat([x, gx, gy], axis=1) y = paddle.concat([x, gx, gy], axis=1)
else: else:
...@@ -273,7 +285,6 @@ class PPYOLOTinyDetBlock(nn.Layer): ...@@ -273,7 +285,6 @@ class PPYOLOTinyDetBlock(nn.Layer):
data_format='NCHW'): data_format='NCHW'):
""" """
PPYOLO Tiny DetBlock layer PPYOLO Tiny DetBlock layer
Args: Args:
ch_in (list): input channel number ch_in (list): input channel number
ch_out (list): output channel number ch_out (list): output channel number
...@@ -333,6 +344,73 @@ class PPYOLOTinyDetBlock(nn.Layer): ...@@ -333,6 +344,73 @@ class PPYOLOTinyDetBlock(nn.Layer):
return route, tip return route, tip
class PPYOLODetBlockCSP(nn.Layer):
def __init__(self,
cfg,
ch_in,
ch_out,
act,
norm_type,
name,
data_format='NCHW'):
"""
PPYOLODetBlockCSP layer
Args:
cfg (list): layer configs for this block
ch_in (int): input channel
ch_out (int): output channel
act (str): default mish
name (str): block name
data_format (str): data format, NCHW or NHWC
"""
super(PPYOLODetBlockCSP, self).__init__()
self.data_format = data_format
self.conv1 = ConvBNLayer(
ch_in,
ch_out,
1,
padding=0,
act=act,
norm_type=norm_type,
name=name + '.left',
data_format=data_format)
self.conv2 = ConvBNLayer(
ch_in,
ch_out,
1,
padding=0,
act=act,
norm_type=norm_type,
name=name + '.right',
data_format=data_format)
self.conv3 = ConvBNLayer(
ch_out * 2,
ch_out * 2,
1,
padding=0,
act=act,
norm_type=norm_type,
name=name,
data_format=data_format)
self.conv_module = nn.Sequential()
for idx, (layer_name, layer, args, kwargs) in enumerate(cfg):
kwargs.update(name=name + layer_name, data_format=data_format)
self.conv_module.add_sublayer(layer_name, layer(*args, **kwargs))
def forward(self, inputs):
conv_left = self.conv1(inputs)
conv_right = self.conv2(inputs)
conv_left = self.conv_module(conv_left)
if self.data_format == 'NCHW':
conv = paddle.concat([conv_left, conv_right], axis=1)
else:
conv = paddle.concat([conv_left, conv_right], axis=-1)
conv = self.conv3(conv)
return conv, conv
@register @register
@serializable @serializable
class YOLOv3FPN(nn.Layer): class YOLOv3FPN(nn.Layer):
...@@ -430,7 +508,12 @@ class PPYOLOFPN(nn.Layer): ...@@ -430,7 +508,12 @@ class PPYOLOFPN(nn.Layer):
in_channels=[512, 1024, 2048], in_channels=[512, 1024, 2048],
norm_type='bn', norm_type='bn',
data_format='NCHW', data_format='NCHW',
**kwargs): coord_conv=False,
conv_block_num=3,
drop_block=False,
block_size=3,
keep_prob=0.9,
spp=False):
""" """
PPYOLOFPN layer PPYOLOFPN layer
...@@ -438,7 +521,12 @@ class PPYOLOFPN(nn.Layer): ...@@ -438,7 +521,12 @@ class PPYOLOFPN(nn.Layer):
in_channels (list): input channels for fpn in_channels (list): input channels for fpn
norm_type (str): batch norm type, default bn norm_type (str): batch norm type, default bn
data_format (str): data format, NCHW or NHWC data_format (str): data format, NCHW or NHWC
kwargs: extra key-value pairs, such as parameter of DropBlock and spp coord_conv (bool): whether use CoordConv or not
conv_block_num (int): conv block num of each pan block
drop_block (bool): whether use DropBlock or not
block_size (int): block size of DropBlock
keep_prob (float): keep probability of DropBlock
spp (bool): whether use spp or not
""" """
super(PPYOLOFPN, self).__init__() super(PPYOLOFPN, self).__init__()
...@@ -446,14 +534,12 @@ class PPYOLOFPN(nn.Layer): ...@@ -446,14 +534,12 @@ class PPYOLOFPN(nn.Layer):
self.in_channels = in_channels self.in_channels = in_channels
self.num_blocks = len(in_channels) self.num_blocks = len(in_channels)
# parse kwargs # parse kwargs
self.coord_conv = kwargs.get('coord_conv', False) self.coord_conv = coord_conv
self.drop_block = kwargs.get('drop_block', False) self.drop_block = drop_block
if self.drop_block: self.block_size = block_size
self.block_size = kwargs.get('block_size', 3) self.keep_prob = keep_prob
self.keep_prob = kwargs.get('keep_prob', 0.9) self.spp = spp
self.conv_block_num = conv_block_num
self.spp = kwargs.get('spp', False)
self.conv_block_num = kwargs.get('conv_block_num', 2)
self.data_format = data_format self.data_format = data_format
if self.coord_conv: if self.coord_conv:
ConvLayer = CoordConv ConvLayer = CoordConv
...@@ -583,14 +669,12 @@ class PPYOLOTinyFPN(nn.Layer): ...@@ -583,14 +669,12 @@ class PPYOLOTinyFPN(nn.Layer):
**kwargs): **kwargs):
""" """
PPYOLO Tiny FPN layer PPYOLO Tiny FPN layer
Args: Args:
in_channels (list): input channels for fpn in_channels (list): input channels for fpn
detection_block_channels (list): channels in fpn detection_block_channels (list): channels in fpn
norm_type (str): batch norm type, default bn norm_type (str): batch norm type, default bn
data_format (str): data format, NCHW or NHWC data_format (str): data format, NCHW or NHWC
kwargs: extra key-value pairs, such as parameter of DropBlock and spp kwargs: extra key-value pairs, such as parameter of DropBlock and spp
""" """
super(PPYOLOTinyFPN, self).__init__() super(PPYOLOTinyFPN, self).__init__()
assert len(in_channels) > 0, "in_channels length should > 0" assert len(in_channels) > 0, "in_channels length should > 0"
...@@ -681,3 +765,197 @@ class PPYOLOTinyFPN(nn.Layer): ...@@ -681,3 +765,197 @@ class PPYOLOTinyFPN(nn.Layer):
@property @property
def out_shape(self): def out_shape(self):
return [ShapeSpec(channels=c) for c in self._out_channels] return [ShapeSpec(channels=c) for c in self._out_channels]
@register
@serializable
class PPYOLOPAN(nn.Layer):
__shared__ = ['norm_type', 'data_format']
def __init__(self,
in_channels=[512, 1024, 2048],
norm_type='bn',
data_format='NCHW',
act='mish',
conv_block_num=3,
drop_block=False,
block_size=3,
keep_prob=0.9,
spp=False):
"""
PPYOLOPAN layer with SPP, DropBlock and CSP connection.
Args:
in_channels (list): input channels for fpn
norm_type (str): batch norm type, default bn
data_format (str): data format, NCHW or NHWC
act (str): activation function, default mish
conv_block_num (int): conv block num of each pan block
drop_block (bool): whether use DropBlock or not
block_size (int): block size of DropBlock
keep_prob (float): keep probability of DropBlock
spp (bool): whether use spp or not
"""
super(PPYOLOPAN, self).__init__()
assert len(in_channels) > 0, "in_channels length should > 0"
self.in_channels = in_channels
self.num_blocks = len(in_channels)
# parse kwargs
self.drop_block = drop_block
self.block_size = block_size
self.keep_prob = keep_prob
self.spp = spp
self.conv_block_num = conv_block_num
self.data_format = data_format
if self.drop_block:
dropblock_cfg = [[
'dropblock', DropBlock, [self.block_size, self.keep_prob],
dict()
]]
else:
dropblock_cfg = []
# fpn
self.fpn_blocks = []
self.fpn_routes = []
fpn_channels = []
for i, ch_in in enumerate(self.in_channels[::-1]):
if i > 0:
ch_in += 512 // (2**(i - 1))
channel = 512 // (2**i)
base_cfg = []
for j in range(self.conv_block_num):
base_cfg += [
# name, layer, args
[
'{}.0'.format(j), ConvBNLayer, [channel, channel, 1],
dict(
padding=0, act=act, norm_type=norm_type)
],
[
'{}.1'.format(j), ConvBNLayer, [channel, channel, 3],
dict(
padding=1, act=act, norm_type=norm_type)
]
]
if i == 0 and self.spp:
base_cfg[3] = [
'spp', SPP, [channel * 4, channel, 1], dict(
pool_size=[5, 9, 13], act=act, norm_type=norm_type)
]
cfg = base_cfg[:4] + dropblock_cfg + base_cfg[4:]
name = 'fpn.{}'.format(i)
fpn_block = self.add_sublayer(
name,
PPYOLODetBlockCSP(cfg, ch_in, channel, act, norm_type, name,
data_format))
self.fpn_blocks.append(fpn_block)
fpn_channels.append(channel * 2)
if i < self.num_blocks - 1:
name = 'fpn_transition.{}'.format(i)
route = self.add_sublayer(
name,
ConvBNLayer(
ch_in=channel * 2,
ch_out=channel,
filter_size=1,
stride=1,
padding=0,
act=act,
norm_type=norm_type,
data_format=data_format,
name=name))
self.fpn_routes.append(route)
# pan
self.pan_blocks = []
self.pan_routes = []
self._out_channels = [512 // (2**(self.num_blocks - 2)), ]
for i in reversed(range(self.num_blocks - 1)):
name = 'pan_transition.{}'.format(i)
route = self.add_sublayer(
name,
ConvBNLayer(
ch_in=fpn_channels[i + 1],
ch_out=fpn_channels[i + 1],
filter_size=3,
stride=2,
padding=1,
act=act,
norm_type=norm_type,
data_format=data_format,
name=name))
self.pan_routes = [route, ] + self.pan_routes
base_cfg = []
ch_in = fpn_channels[i] + fpn_channels[i + 1]
channel = 512 // (2**i)
for j in range(self.conv_block_num):
base_cfg += [
# name, layer, args
[
'{}.0'.format(j), ConvBNLayer, [channel, channel, 1],
dict(
padding=0, act=act, norm_type=norm_type)
],
[
'{}.1'.format(j), ConvBNLayer, [channel, channel, 3],
dict(
padding=1, act=act, norm_type=norm_type)
]
]
cfg = base_cfg[:4] + dropblock_cfg + base_cfg[4:]
name = 'pan.{}'.format(i)
pan_block = self.add_sublayer(
name,
PPYOLODetBlockCSP(cfg, ch_in, channel, act, norm_type, name,
data_format))
self.pan_blocks = [pan_block, ] + self.pan_blocks
self._out_channels.append(channel * 2)
self._out_channels = self._out_channels[::-1]
def forward(self, blocks):
assert len(blocks) == self.num_blocks
blocks = blocks[::-1]
# fpn
fpn_feats = []
for i, block in enumerate(blocks):
if i > 0:
if self.data_format == 'NCHW':
block = paddle.concat([route, block], axis=1)
else:
block = paddle.concat([route, block], axis=-1)
route, tip = self.fpn_blocks[i](block)
fpn_feats.append(tip)
if i < self.num_blocks - 1:
route = self.fpn_routes[i](route)
route = F.interpolate(
route, scale_factor=2., data_format=self.data_format)
pan_feats = [fpn_feats[-1], ]
route = fpn_feats[self.num_blocks - 1]
for i in reversed(range(self.num_blocks - 1)):
block = fpn_feats[i]
route = self.pan_routes[i](route)
if self.data_format == 'NCHW':
block = paddle.concat([route, block], axis=1)
else:
block = paddle.concat([route, block], axis=-1)
route, tip = self.pan_blocks[i](block)
pan_feats.append(tip)
return pan_feats[::-1]
@classmethod
def from_config(cls, cfg, input_shape):
return {'in_channels': [i.channels for i in input_shape], }
@property
def out_shape(self):
return [ShapeSpec(channels=c) for c in self._out_channels]
...@@ -41,9 +41,14 @@ __all__ = [ ...@@ -41,9 +41,14 @@ __all__ = [
'collect_fpn_proposals', 'collect_fpn_proposals',
'matrix_nms', 'matrix_nms',
'batch_norm', 'batch_norm',
'mish',
] ]
def mish(x):
return x * paddle.tanh(F.softplus(x))
def batch_norm(ch, def batch_norm(ch,
norm_type='bn', norm_type='bn',
norm_decay=0., norm_decay=0.,
......
...@@ -38,21 +38,24 @@ PP-YOLO improved performance and speed of YOLOv3 with following methods: ...@@ -38,21 +38,24 @@ PP-YOLO improved performance and speed of YOLOv3 with following methods:
| Model | GPU number | images/GPU | backbone | input shape | Box AP<sup>val</sup> | Box AP<sup>test</sup> | V100 FP32(FPS) | V100 TensorRT FP16(FPS) | download | config | | Model | GPU number | images/GPU | backbone | input shape | Box AP<sup>val</sup> | Box AP<sup>test</sup> | V100 FP32(FPS) | V100 TensorRT FP16(FPS) | download | config |
|:------------------------:|:----------:|:----------:|:----------:| :----------:| :------------------: | :-------------------: | :------------: | :---------------------: | :------: | :-----: | |:------------------------:|:----------:|:----------:|:----------:| :----------:| :------------------: | :-------------------: | :------------: | :---------------------: | :------: | :-----: |
| YOLOv4(AlexyAB) | - | - | CSPDarknet | 608 | - | 43.5 | 62 | 105.5 | [model](https://paddlemodels.bj.bcebos.com/object_detection/yolov4_cspdarknet.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/yolov4/yolov4_csdarknet.yml) | | YOLOv4(AlexyAB) | - | - | CSPDarknet | 608 | - | 43.5 | 62 | 105.5 | [model](https://paddlemodels.bj.bcebos.com/object_detection/yolov4_cspdarknet.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/yolov4/yolov4_csdarknet.yml) |
| YOLOv4(AlexyAB) | - | - | CSPDarknet | 512 | - | 43.0 | 83 | 138.4 | [model](https://paddlemodels.bj.bcebos.com/object_detection/yolov4_cspdarknet.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/yolov4/yolov4_csdarknet.yml) | | YOLOv4(AlexyAB) | - | - | CSPDarknet | 512 | - | 43.0 | 83 | 138.4 | [model](https://paddlemodels.bj.bcebos.com/object_detection/yolov4_cspdarknet.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/yolov4/yolov4_csdarknet.yml) |
| YOLOv4(AlexyAB) | - | - | CSPDarknet | 416 | - | 41.2 | 96 | 164.0 | [model](https://paddlemodels.bj.bcebos.com/object_detection/yolov4_cspdarknet.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/yolov4/yolov4_csdarknet.yml) | | YOLOv4(AlexyAB) | - | - | CSPDarknet | 416 | - | 41.2 | 96 | 164.0 | [model](https://paddlemodels.bj.bcebos.com/object_detection/yolov4_cspdarknet.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/yolov4/yolov4_csdarknet.yml) |
| YOLOv4(AlexyAB) | - | - | CSPDarknet | 320 | - | 38.0 | 123 | 199.0 | [model](https://paddlemodels.bj.bcebos.com/object_detection/yolov4_cspdarknet.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/yolov4/yolov4_csdarknet.yml) | | YOLOv4(AlexyAB) | - | - | CSPDarknet | 320 | - | 38.0 | 123 | 199.0 | [model](https://paddlemodels.bj.bcebos.com/object_detection/yolov4_cspdarknet.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/yolov4/yolov4_csdarknet.yml) |
| PP-YOLO | 8 | 24 | ResNet50vd | 608 | 44.8 | 45.2 | 72.9 | 155.6 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo.yml) | | PP-YOLO | 8 | 24 | ResNet50vd | 608 | 44.8 | 45.2 | 72.9 | 155.6 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo.yml) |
| PP-YOLO | 8 | 24 | ResNet50vd | 512 | 43.9 | 44.4 | 89.9 | 188.4 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo.yml) | | PP-YOLO | 8 | 24 | ResNet50vd | 512 | 43.9 | 44.4 | 89.9 | 188.4 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo.yml) |
| PP-YOLO | 8 | 24 | ResNet50vd | 416 | 42.1 | 42.5 | 109.1 | 215.4 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo.yml) | | PP-YOLO | 8 | 24 | ResNet50vd | 416 | 42.1 | 42.5 | 109.1 | 215.4 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo.yml) |
| PP-YOLO | 8 | 24 | ResNet50vd | 320 | 38.9 | 39.3 | 132.2 | 242.2 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo.yml) | | PP-YOLO | 8 | 24 | ResNet50vd | 320 | 38.9 | 39.3 | 132.2 | 242.2 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo.yml) |
| PP-YOLO_2x | 8 | 24 | ResNet50vd | 608 | 45.3 | 45.9 | 72.9 | 155.6 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_2x.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo_2x.yml) | | PP-YOLO_2x | 8 | 24 | ResNet50vd | 608 | 45.3 | 45.9 | 72.9 | 155.6 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_2x.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo_2x.yml) |
| PP-YOLO_2x | 8 | 24 | ResNet50vd | 512 | 44.4 | 45.0 | 89.9 | 188.4 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_2x.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo_2x.yml) | | PP-YOLO_2x | 8 | 24 | ResNet50vd | 512 | 44.4 | 45.0 | 89.9 | 188.4 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_2x.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo_2x.yml) |
| PP-YOLO_2x | 8 | 24 | ResNet50vd | 416 | 42.7 | 43.2 | 109.1 | 215.4 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_2x.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo_2x.yml) | | PP-YOLO_2x | 8 | 24 | ResNet50vd | 416 | 42.7 | 43.2 | 109.1 | 215.4 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_2x.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo_2x.yml) |
| PP-YOLO_2x | 8 | 24 | ResNet50vd | 320 | 39.5 | 40.1 | 132.2 | 242.2 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_2x.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo_2x.yml) | | PP-YOLO_2x | 8 | 24 | ResNet50vd | 320 | 39.5 | 40.1 | 132.2 | 242.2 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_2x.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo_2x.yml) |
| PP-YOLO_ResNet18vd | 4 | 32 | ResNet18vd | 512 | 29.3 | 29.5 | 357.1 | 657.9 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_r18vd.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo_r18vd.yml) | | PP-YOLO | 4 | 32 | ResNet18vd | 512 | 29.3 | 29.5 | 357.1 | 657.9 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_r18vd.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo_r18vd.yml) |
| PP-YOLO_ResNet18vd | 4 | 32 | ResNet18vd | 416 | 28.6 | 28.9 | 409.8 | 719.4 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_r18vd.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo_r18vd.yml) | | PP-YOLO | 4 | 32 | ResNet18vd | 416 | 28.6 | 28.9 | 409.8 | 719.4 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_r18vd.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo_r18vd.yml) |
| PP-YOLO_ResNet18vd | 4 | 32 | ResNet18vd | 320 | 26.2 | 26.4 | 480.7 | 763.4 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_r18vd.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo_r18vd.yml) | | PP-YOLO | 4 | 32 | ResNet18vd | 320 | 26.2 | 26.4 | 480.7 | 763.4 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_r18vd.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo_r18vd.yml) |
| PP-YOLOv2 | 8 | 12 | ResNet50vd | 640 | 49.1 | 49.5 | - | - | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolov2_r50vd_dcn.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolov2_r50vd_dcn.yml) |
| PP-YOLOv2 | 8 | 12 | ResNet101vd | 640 | 49.7 | 50.1 | - | - | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolov2_r101vd_dcn.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolov2_r101vd_dcn.yml) |
**Notes:** **Notes:**
...@@ -69,8 +72,8 @@ PP-YOLO improved performance and speed of YOLOv3 with following methods: ...@@ -69,8 +72,8 @@ PP-YOLO improved performance and speed of YOLOv3 with following methods:
| Model | GPU number | images/GPU | Model Size | input shape | Box AP<sup>val</sup> | Box AP50<sup>val</sup> | Kirin 990 1xCore(FPS) | download | inference model download | config | | Model | GPU number | images/GPU | Model Size | input shape | Box AP<sup>val</sup> | Box AP50<sup>val</sup> | Kirin 990 1xCore(FPS) | download | inference model download | config |
|:----------------------------:|:----------:|:----------:| :--------: | :----------:| :------------------: | :--------------------: | :-------------------: | :------: | :----------------------: | :-----: | |:----------------------------:|:----------:|:----------:| :--------: | :----------:| :------------------: | :--------------------: | :-------------------: | :------: | :----------------------: | :-----: |
| PP-YOLO_MobileNetV3_large | 4 | 32 | 18MB | 320 | 23.2 | 42.6 | 15.6 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_mobilenet_v3_large.pdparams) | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_mobilenet_v3_large.tar) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo_mobilenet_v3_large.yml) | | PP-YOLO_MobileNetV3_large | 4 | 32 | 18MB | 320 | 23.2 | 42.6 | 15.6 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_mobilenet_v3_large.pdparams) | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_mobilenet_v3_large.tar) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo_mobilenet_v3_large.yml) |
| PP-YOLO_MobileNetV3_small | 4 | 32 | 11MB | 320 | 17.2 | 33.8 | 28.6 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_mobilenet_v3_small.pdparams) | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_mobilenet_v3_small.tar) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo_mobilenet_v3_small.yml) | | PP-YOLO_MobileNetV3_small | 4 | 32 | 11MB | 320 | 17.2 | 33.8 | 28.6 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_mobilenet_v3_small.pdparams) | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_mobilenet_v3_small.tar) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo_mobilenet_v3_small.yml) |
**Notes:** **Notes:**
...@@ -82,7 +85,7 @@ PP-YOLO improved performance and speed of YOLOv3 with following methods: ...@@ -82,7 +85,7 @@ PP-YOLO improved performance and speed of YOLOv3 with following methods:
| Model | GPU number | images/GPU | Prune Ratio | Teacher Model | Model Size | input shape | Box AP<sup>val</sup> | Kirin 990 1xCore(FPS) | download | inference model download | config | | Model | GPU number | images/GPU | Prune Ratio | Teacher Model | Model Size | input shape | Box AP<sup>val</sup> | Kirin 990 1xCore(FPS) | download | inference model download | config |
|:----------------------------:|:----------:|:----------:| :---------: | :-----------------------: | :--------: | :----------:| :------------------: | :-------------------: | :------: | :----------------------: | :-----: | |:----------------------------:|:----------:|:----------:| :---------: | :-----------------------: | :--------: | :----------:| :------------------: | :-------------------: | :------: | :----------------------: | :-----: |
| PP-YOLO_MobileNetV3_small | 4 | 32 | 75% | PP-YOLO_MobileNetV3_large | 4.2MB | 320 | 16.2 | 39.8 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_mobilenet_v3_small_prune75_distillby_mobilenet_v3_large.pdparams) | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_mobilenet_v3_small_prune75_distillby_mobilenet_v3_large.tar) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo_mobilenet_v3_small.yml) | | PP-YOLO_MobileNetV3_small | 4 | 32 | 75% | PP-YOLO_MobileNetV3_large | 4.2MB | 320 | 16.2 | 39.8 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_mobilenet_v3_small_prune75_distillby_mobilenet_v3_large.pdparams) | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_mobilenet_v3_small_prune75_distillby_mobilenet_v3_large.tar) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo_mobilenet_v3_small.yml) |
- Slim PP-YOLO is trained by slim traing method from [Distill pruned model](../../slim/extentions/distill_pruned_model/README.md),distill training pruned PP-YOLO_MobileNetV3_small model with PP-YOLO_MobileNetV3_large model as the teacher model - Slim PP-YOLO is trained by slim traing method from [Distill pruned model](../../slim/extentions/distill_pruned_model/README.md),distill training pruned PP-YOLO_MobileNetV3_small model with PP-YOLO_MobileNetV3_large model as the teacher model
- Pruning detectiom head of PP-YOLO model with ratio as 75%, while the arguments are `--pruned_params="yolo_block.0.2.conv.weights,yolo_block.0.tip.conv.weights,yolo_block.1.2.conv.weights,yolo_block.1.tip.conv.weights" --pruned_ratios="0.75,0.75,0.75,0.75"` - Pruning detectiom head of PP-YOLO model with ratio as 75%, while the arguments are `--pruned_params="yolo_block.0.2.conv.weights,yolo_block.0.tip.conv.weights,yolo_block.1.2.conv.weights,yolo_block.1.tip.conv.weights" --pruned_ratios="0.75,0.75,0.75,0.75"`
...@@ -108,9 +111,9 @@ PP-YOLO trained on Pascal VOC dataset as follows: ...@@ -108,9 +111,9 @@ PP-YOLO trained on Pascal VOC dataset as follows:
| Model | GPU number | images/GPU | backbone | input shape | Box AP50<sup>val</sup> | download | config | | Model | GPU number | images/GPU | backbone | input shape | Box AP50<sup>val</sup> | download | config |
|:------------------:|:----------:|:----------:|:----------:| :----------:| :--------------------: | :------: | :-----: | |:------------------:|:----------:|:----------:|:----------:| :----------:| :--------------------: | :------: | :-----: |
| PP-YOLO | 8 | 12 | ResNet50vd | 608 | 84.9 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_voc.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo_voc.yml) | | PP-YOLO | 8 | 12 | ResNet50vd | 608 | 84.9 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_voc.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo_voc.yml) |
| PP-YOLO | 8 | 12 | ResNet50vd | 416 | 84.3 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_voc.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo_voc.yml) | | PP-YOLO | 8 | 12 | ResNet50vd | 416 | 84.3 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_voc.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo_voc.yml) |
| PP-YOLO | 8 | 12 | ResNet50vd | 320 | 82.2 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_voc.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo_voc.yml) | | PP-YOLO | 8 | 12 | ResNet50vd | 320 | 82.2 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_voc.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo_voc.yml) |
## Getting Start ## Getting Start
......
...@@ -38,21 +38,23 @@ PP-YOLO从如下方面优化和提升YOLOv3模型的精度和速度: ...@@ -38,21 +38,23 @@ PP-YOLO从如下方面优化和提升YOLOv3模型的精度和速度:
| 模型 | GPU个数 | 每GPU图片个数 | 骨干网络 | 输入尺寸 | Box AP<sup>val</sup> | Box AP<sup>test</sup> | V100 FP32(FPS) | V100 TensorRT FP16(FPS) | 模型下载 | 配置文件 | | 模型 | GPU个数 | 每GPU图片个数 | 骨干网络 | 输入尺寸 | Box AP<sup>val</sup> | Box AP<sup>test</sup> | V100 FP32(FPS) | V100 TensorRT FP16(FPS) | 模型下载 | 配置文件 |
|:------------------------:|:-------:|:-------------:|:----------:| :-------:| :------------------: | :-------------------: | :------------: | :---------------------: | :------: | :------: | |:------------------------:|:-------:|:-------------:|:----------:| :-------:| :------------------: | :-------------------: | :------------: | :---------------------: | :------: | :------: |
| YOLOv4(AlexyAB) | - | - | CSPDarknet | 608 | - | 43.5 | 62 | 105.5 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/yolov4_cspdarknet.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/yolov4/yolov4_csdarknet.yml) | | YOLOv4(AlexyAB) | - | - | CSPDarknet | 608 | - | 43.5 | 62 | 105.5 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/yolov4_cspdarknet.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/yolov4/yolov4_csdarknet.yml) |
| YOLOv4(AlexyAB) | - | - | CSPDarknet | 512 | - | 43.0 | 83 | 138.4 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/yolov4_cspdarknet.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/yolov4/yolov4_csdarknet.yml) | | YOLOv4(AlexyAB) | - | - | CSPDarknet | 512 | - | 43.0 | 83 | 138.4 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/yolov4_cspdarknet.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/yolov4/yolov4_csdarknet.yml) |
| YOLOv4(AlexyAB) | - | - | CSPDarknet | 416 | - | 41.2 | 96 | 164.0 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/yolov4_cspdarknet.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/yolov4/yolov4_csdarknet.yml) | | YOLOv4(AlexyAB) | - | - | CSPDarknet | 416 | - | 41.2 | 96 | 164.0 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/yolov4_cspdarknet.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/yolov4/yolov4_csdarknet.yml) |
| YOLOv4(AlexyAB) | - | - | CSPDarknet | 320 | - | 38.0 | 123 | 199.0 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/yolov4_cspdarknet.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/yolov4/yolov4_csdarknet.yml) | | YOLOv4(AlexyAB) | - | - | CSPDarknet | 320 | - | 38.0 | 123 | 199.0 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/yolov4_cspdarknet.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/yolov4/yolov4_csdarknet.yml) |
| PP-YOLO | 8 | 24 | ResNet50vd | 608 | 44.8 | 45.2 | 72.9 | 155.6 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo.yml) | | PP-YOLO | 8 | 24 | ResNet50vd | 608 | 44.8 | 45.2 | 72.9 | 155.6 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo.yml) |
| PP-YOLO | 8 | 24 | ResNet50vd | 512 | 43.9 | 44.4 | 89.9 | 188.4 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo.yml) | | PP-YOLO | 8 | 24 | ResNet50vd | 512 | 43.9 | 44.4 | 89.9 | 188.4 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo.yml) |
| PP-YOLO | 8 | 24 | ResNet50vd | 416 | 42.1 | 42.5 | 109.1 | 215.4 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo.yml) | | PP-YOLO | 8 | 24 | ResNet50vd | 416 | 42.1 | 42.5 | 109.1 | 215.4 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo.yml) |
| PP-YOLO | 8 | 24 | ResNet50vd | 320 | 38.9 | 39.3 | 132.2 | 242.2 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo.yml) | | PP-YOLO | 8 | 24 | ResNet50vd | 320 | 38.9 | 39.3 | 132.2 | 242.2 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo.yml) |
| PP-YOLO_2x | 8 | 24 | ResNet50vd | 608 | 45.3 | 45.9 | 72.9 | 155.6 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo.yml) | | PP-YOLO_2x | 8 | 24 | ResNet50vd | 608 | 45.3 | 45.9 | 72.9 | 155.6 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo.yml) |
| PP-YOLO_2x | 8 | 24 | ResNet50vd | 512 | 44.4 | 45.0 | 89.9 | 188.4 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo.yml) | | PP-YOLO_2x | 8 | 24 | ResNet50vd | 512 | 44.4 | 45.0 | 89.9 | 188.4 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo.yml) |
| PP-YOLO_2x | 8 | 24 | ResNet50vd | 416 | 42.7 | 43.2 | 109.1 | 215.4 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo.yml) | | PP-YOLO_2x | 8 | 24 | ResNet50vd | 416 | 42.7 | 43.2 | 109.1 | 215.4 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo.yml) |
| PP-YOLO_2x | 8 | 24 | ResNet50vd | 320 | 39.5 | 40.1 | 132.2 | 242.2 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo.yml) | | PP-YOLO_2x | 8 | 24 | ResNet50vd | 320 | 39.5 | 40.1 | 132.2 | 242.2 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo.yml) |
| PP-YOLO_ResNet18vd | 4 | 32 | ResNet18vd | 512 | 29.3 | 29.5 | 357.1 | 657.9 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_r18vd.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo_r18vd.yml) | | PP-YOLO | 4 | 32 | ResNet18vd | 512 | 29.3 | 29.5 | 357.1 | 657.9 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_r18vd.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo_r18vd.yml) |
| PP-YOLO_ResNet18vd | 4 | 32 | ResNet18vd | 416 | 28.6 | 28.9 | 409.8 | 719.4 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_r18vd.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo_r18vd.yml) | | PP-YOLO | 4 | 32 | ResNet18vd | 416 | 28.6 | 28.9 | 409.8 | 719.4 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_r18vd.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo_r18vd.yml) |
| PP-YOLO_ResNet18vd | 4 | 32 | ResNet18vd | 320 | 26.2 | 26.4 | 480.7 | 763.4 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_r18vd.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo_r18vd.yml) | | PP-YOLO | 4 | 32 | ResNet18vd | 320 | 26.2 | 26.4 | 480.7 | 763.4 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_r18vd.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo_r18vd.yml) |
| PP-YOLOv2 | 8 | 12 | ResNet50vd | 640 | 49.1 | 49.5 | - | - | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolov2_r50vd_dcn.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolov2_r50vd_dcn.yml) |
| PP-YOLOv2 | 8 | 12 | ResNet101vd | 640 | 49.7 | 50.1 | - | - | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolov2_r101vd_dcn.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolov2_r101vd_dcn.yml) |
**注意:** **注意:**
...@@ -70,8 +72,8 @@ PP-YOLO从如下方面优化和提升YOLOv3模型的精度和速度: ...@@ -70,8 +72,8 @@ PP-YOLO从如下方面优化和提升YOLOv3模型的精度和速度:
| 模型 | GPU个数 | 每GPU图片个数 | 模型体积 | 输入尺寸 | Box AP<sup>val</sup> | Box AP50<sup>val</sup> | Kirin 990 1xCore (FPS) | 模型下载 | 预测模型下载 | 配置文件 | | 模型 | GPU个数 | 每GPU图片个数 | 模型体积 | 输入尺寸 | Box AP<sup>val</sup> | Box AP50<sup>val</sup> | Kirin 990 1xCore (FPS) | 模型下载 | 预测模型下载 | 配置文件 |
|:----------------------------:|:-------:|:-------------:|:----------:| :-------:| :------------------: | :--------------------: | :--------------------: | :------: | :----------: | :------: | |:----------------------------:|:-------:|:-------------:|:----------:| :-------:| :------------------: | :--------------------: | :--------------------: | :------: | :----------: | :------: |
| PP-YOLO_MobileNetV3_large | 4 | 32 | 18MB | 320 | 23.2 | 42.6 | 14.1 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_mobilenet_v3_large.pdparams) | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_mobilenet_v3_large.tar) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo_mobilenet_v3_large.yml) | | PP-YOLO_MobileNetV3_large | 4 | 32 | 18MB | 320 | 23.2 | 42.6 | 14.1 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_mobilenet_v3_large.pdparams) | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_mobilenet_v3_large.tar) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo_mobilenet_v3_large.yml) |
| PP-YOLO_MobileNetV3_small | 4 | 32 | 11MB | 320 | 17.2 | 33.8 | 21.5 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_mobilenet_v3_small.pdparams) | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_mobilenet_v3_large.tar) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo_mobilenet_v3_small.yml) | | PP-YOLO_MobileNetV3_small | 4 | 32 | 11MB | 320 | 17.2 | 33.8 | 21.5 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_mobilenet_v3_small.pdparams) | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_mobilenet_v3_large.tar) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo_mobilenet_v3_small.yml) |
- PP-YOLO_MobileNetV3 模型使用COCO数据集中train2017作为训练集,使用val2017作为测试集,Box AP<sup>val</sup>`mAP(IoU=0.5:0.95)`评估结果, Box AP50<sup>val</sup>`mAP(IoU=0.5)`评估结果。 - PP-YOLO_MobileNetV3 模型使用COCO数据集中train2017作为训练集,使用val2017作为测试集,Box AP<sup>val</sup>`mAP(IoU=0.5:0.95)`评估结果, Box AP50<sup>val</sup>`mAP(IoU=0.5)`评估结果。
- PP-YOLO_MobileNetV3 模型训练过程中使用4GPU,每GPU batch size为32进行训练,如训练GPU数和batch size不使用上述配置,须参考[FAQ](../../docs/FAQ.md)调整学习率和迭代次数。 - PP-YOLO_MobileNetV3 模型训练过程中使用4GPU,每GPU batch size为32进行训练,如训练GPU数和batch size不使用上述配置,须参考[FAQ](../../docs/FAQ.md)调整学习率和迭代次数。
...@@ -81,7 +83,7 @@ PP-YOLO从如下方面优化和提升YOLOv3模型的精度和速度: ...@@ -81,7 +83,7 @@ PP-YOLO从如下方面优化和提升YOLOv3模型的精度和速度:
| 模型 | GPU 个数 | 每GPU图片个数 | 裁剪率 | Teacher模型 | 模型体积 | 输入尺寸 | Box AP<sup>val</sup> | Kirin 990 1xCore (FPS) | 模型下载 | 预测模型下载 | 配置文件 | | 模型 | GPU 个数 | 每GPU图片个数 | 裁剪率 | Teacher模型 | 模型体积 | 输入尺寸 | Box AP<sup>val</sup> | Kirin 990 1xCore (FPS) | 模型下载 | 预测模型下载 | 配置文件 |
|:----------------------------:|:----------:|:-------------:| :---------: | :-----------------------: | :--------: | :----------:| :------------------: | :--------------------: | :------: | :----------: | :------: | |:----------------------------:|:----------:|:-------------:| :---------: | :-----------------------: | :--------: | :----------:| :------------------: | :--------------------: | :------: | :----------: | :------: |
| PP-YOLO_MobileNetV3_small | 4 | 32 | 75% | PP-YOLO_MobileNetV3_large | 4.2MB | 320 | 16.2 | 39.8 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_mobilenet_v3_small_prune75_distillby_mobilenet_v3_large.pdparams) | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_mobilenet_v3_small_prune75_distillby_mobilenet_v3_large.tar) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo_mobilenet_v3_small.yml) | | PP-YOLO_MobileNetV3_small | 4 | 32 | 75% | PP-YOLO_MobileNetV3_large | 4.2MB | 320 | 16.2 | 39.8 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_mobilenet_v3_small_prune75_distillby_mobilenet_v3_large.pdparams) | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_mobilenet_v3_small_prune75_distillby_mobilenet_v3_large.tar) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo_mobilenet_v3_small.yml) |
- PP-YOLO 轻量级裁剪模型采用[蒸馏通道剪裁模型](../../slim/extentions/distill_pruned_model/README.md) 的方式训练得到,基于 PP-YOLO_MobileNetV3_small 模型对Head部分做卷积通道剪裁后使用 PP-YOLO_MobileNetV3_large 模型进行蒸馏训练 - PP-YOLO 轻量级裁剪模型采用[蒸馏通道剪裁模型](../../slim/extentions/distill_pruned_model/README.md) 的方式训练得到,基于 PP-YOLO_MobileNetV3_small 模型对Head部分做卷积通道剪裁后使用 PP-YOLO_MobileNetV3_large 模型进行蒸馏训练
- 卷积通道检测对Head部分剪裁掉75%的通道数,及剪裁参数为`--pruned_params="yolo_block.0.2.conv.weights,yolo_block.0.tip.conv.weights,yolo_block.1.2.conv.weights,yolo_block.1.tip.conv.weights" --pruned_ratios="0.75,0.75,0.75,0.75"` - 卷积通道检测对Head部分剪裁掉75%的通道数,及剪裁参数为`--pruned_params="yolo_block.0.2.conv.weights,yolo_block.0.tip.conv.weights,yolo_block.1.2.conv.weights,yolo_block.1.tip.conv.weights" --pruned_ratios="0.75,0.75,0.75,0.75"`
...@@ -105,9 +107,9 @@ PP-YOLO在Pascal VOC数据集上训练模型如下: ...@@ -105,9 +107,9 @@ PP-YOLO在Pascal VOC数据集上训练模型如下:
| 模型 | GPU个数 | 每GPU图片个数 | 骨干网络 | 输入尺寸 | Box AP50<sup>val</sup> | 模型下载 | 配置文件 | | 模型 | GPU个数 | 每GPU图片个数 | 骨干网络 | 输入尺寸 | Box AP50<sup>val</sup> | 模型下载 | 配置文件 |
|:------------------:|:-------:|:-------------:|:----------:| :----------:| :--------------------: | :------: | :-----: | |:------------------:|:-------:|:-------------:|:----------:| :----------:| :--------------------: | :------: | :-----: |
| PP-YOLO | 8 | 12 | ResNet50vd | 608 | 84.9 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_voc.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo_voc.yml) | | PP-YOLO | 8 | 12 | ResNet50vd | 608 | 84.9 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_voc.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo_voc.yml) |
| PP-YOLO | 8 | 12 | ResNet50vd | 416 | 84.3 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_voc.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo_voc.yml) | | PP-YOLO | 8 | 12 | ResNet50vd | 416 | 84.3 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_voc.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo_voc.yml) |
| PP-YOLO | 8 | 12 | ResNet50vd | 320 | 82.2 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_voc.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/ppyolo/ppyolo_voc.yml) | | PP-YOLO | 8 | 12 | ResNet50vd | 320 | 82.2 | [model](https://paddlemodels.bj.bcebos.com/object_detection/ppyolo_voc.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/static/configs/ppyolo/ppyolo_voc.yml) |
## 使用说明 ## 使用说明
......
architecture: YOLOv3
use_gpu: true
max_iters: 450000
log_iter: 100
save_dir: output
snapshot_iter: 10000
metric: COCO
pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/ResNet101_vd_ssld_pretrained.tar
weights: output/ppyolov2_r101vd_dcn/model_final
num_classes: 80
use_fine_grained_loss: true
use_ema: true
ema_decay: 0.9998
YOLOv3:
backbone: ResNet
yolo_head: YOLOv3PANHead
use_fine_grained_loss: true
ResNet:
norm_type: sync_bn
freeze_at: 0
freeze_norm: false
norm_decay: 0.
depth: 101
feature_maps: [3, 4, 5]
variant: d
dcn_v2_stages: [5]
YOLOv3PANHead:
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]]
norm_decay: 0.
iou_aware: true
iou_aware_factor: 0.5
scale_x_y: 1.05
spp: true
yolo_loss: YOLOv3Loss
nms: MatrixNMS
drop_block: true
YOLOv3Loss:
ignore_thresh: 0.7
scale_x_y: 1.05
label_smooth: false
use_fine_grained_loss: true
iou_loss: IouLoss
iou_aware_loss: IouAwareLoss
IouLoss:
loss_weight: 2.5
max_height: 768
max_width: 768
IouAwareLoss:
loss_weight: 1.0
max_height: 768
max_width: 768
MatrixNMS:
background_label: -1
keep_top_k: 100
normalized: false
score_threshold: 0.01
post_threshold: 0.01
LearningRate:
base_lr: 0.005
schedulers:
- !PiecewiseDecay
gamma: 0.1
milestones:
- 300000
- !LinearWarmup
start_factor: 0.
steps: 4000
OptimizerBuilder:
clip_grad_by_norm: 35.
optimizer:
momentum: 0.9
type: Momentum
regularizer:
factor: 0.0005
type: L2
_READER_: 'ppyolov2_reader.yml'
architecture: YOLOv3
use_gpu: true
max_iters: 450000
log_iter: 100
save_dir: output
snapshot_iter: 10000
metric: COCO
pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_vd_ssld_pretrained.tar
weights: output/ppyolov2_r50vd_dcn/model_final
num_classes: 80
use_fine_grained_loss: true
use_ema: true
ema_decay: 0.9998
YOLOv3:
backbone: ResNet
yolo_head: YOLOv3PANHead
use_fine_grained_loss: true
ResNet:
norm_type: sync_bn
freeze_at: 0
freeze_norm: false
norm_decay: 0.
depth: 50
feature_maps: [3, 4, 5]
variant: d
dcn_v2_stages: [5]
YOLOv3PANHead:
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]]
norm_decay: 0.
iou_aware: true
iou_aware_factor: 0.5
scale_x_y: 1.05
spp: true
yolo_loss: YOLOv3Loss
nms: MatrixNMS
drop_block: true
YOLOv3Loss:
ignore_thresh: 0.7
scale_x_y: 1.05
label_smooth: false
use_fine_grained_loss: true
iou_loss: IouLoss
iou_aware_loss: IouAwareLoss
IouLoss:
loss_weight: 2.5
max_height: 768
max_width: 768
IouAwareLoss:
loss_weight: 1.0
max_height: 768
max_width: 768
MatrixNMS:
background_label: -1
keep_top_k: 100
normalized: false
score_threshold: 0.01
post_threshold: 0.01
LearningRate:
base_lr: 0.005
schedulers:
- !PiecewiseDecay
gamma: 0.1
milestones:
- 300000
- !LinearWarmup
start_factor: 0.
steps: 4000
OptimizerBuilder:
clip_grad_by_norm: 35.
optimizer:
momentum: 0.9
type: Momentum
regularizer:
factor: 0.0005
type: L2
_READER_: 'ppyolov2_reader.yml'
TrainReader:
inputs_def:
fields: ['image', 'gt_bbox', 'gt_class', 'gt_score']
num_max_boxes: 100
dataset:
!COCODataSet
image_dir: train2017
anno_path: annotations/instances_train2017.json
dataset_dir: dataset/coco
with_background: false
sample_transforms:
- !DecodeImage
to_rgb: True
with_mixup: True
- !MixupImage
alpha: 1.5
beta: 1.5
- !ColorDistort {}
- !RandomExpand
ratio: 2.0
fill_value: [123.675, 116.28, 103.53]
- !RandomCrop {}
- !RandomFlipImage
is_normalized: false
- !NormalizeBox {}
- !PadBox
num_max_boxes: 100
- !BboxXYXY2XYWH {}
batch_transforms:
- !RandomShape
sizes: [320, 352, 384, 416, 448, 480, 512, 544, 576, 608, 640, 672, 704, 736, 768]
random_inter: True
- !NormalizeImage
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
is_scale: True
is_channel_first: false
- !Permute
to_bgr: false
channel_first: True
# Gt2YoloTarget is only used when use_fine_grained_loss set as true,
# this operator will be deleted automatically if use_fine_grained_loss
# is set as false
- !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]
batch_size: 12
shuffle: true
mixup_epoch: 25000
drop_last: true
worker_num: 8
bufsize: 4
use_process: true
EvalReader:
inputs_def:
fields: ['image', 'im_size', 'im_id']
num_max_boxes: 100
dataset:
!COCODataSet
image_dir: val2017
anno_path: annotations/instances_val2017.json
dataset_dir: dataset/coco
with_background: false
sample_transforms:
- !DecodeImage
to_rgb: True
- !ResizeImage
target_size: 640
interp: 2
- !NormalizeImage
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
is_scale: True
is_channel_first: false
- !PadBox
num_max_boxes: 50
- !Permute
to_bgr: false
channel_first: True
batch_size: 8
drop_empty: false
worker_num: 8
bufsize: 4
TestReader:
inputs_def:
image_shape: [3, 640, 640]
fields: ['image', 'im_size', 'im_id']
dataset:
!ImageFolder
anno_path: annotations/instances_val2017.json
with_background: false
sample_transforms:
- !DecodeImage
to_rgb: True
- !ResizeImage
target_size: 640
interp: 2
- !NormalizeImage
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
is_scale: True
is_channel_first: false
- !Permute
to_bgr: false
channel_first: True
batch_size: 1
...@@ -192,6 +192,8 @@ class YOLOv3Head(object): ...@@ -192,6 +192,8 @@ class YOLOv3Head(object):
if act == 'leaky': if act == 'leaky':
out = fluid.layers.leaky_relu(x=out, alpha=0.1) out = fluid.layers.leaky_relu(x=out, alpha=0.1)
elif act == 'mish':
out = fluid.layers.mish(out)
return out return out
def _spp_module(self, input, name=""): def _spp_module(self, input, name=""):
...@@ -657,7 +659,6 @@ class YOLOv4Head(YOLOv3Head): ...@@ -657,7 +659,6 @@ class YOLOv4Head(YOLOv3Head):
class PPYOLOTinyHead(YOLOv3Head): class PPYOLOTinyHead(YOLOv3Head):
""" """
Head block for YOLOv3 network Head block for YOLOv3 network
Args: Args:
norm_decay (float): weight decay for normalization layer weights norm_decay (float): weight decay for normalization layer weights
num_classes (int): number of output classes num_classes (int): number of output classes
...@@ -781,11 +782,9 @@ class PPYOLOTinyHead(YOLOv3Head): ...@@ -781,11 +782,9 @@ class PPYOLOTinyHead(YOLOv3Head):
def _get_outputs(self, input, is_train=True): def _get_outputs(self, input, is_train=True):
""" """
Get PP-YOLO tiny head output Get PP-YOLO tiny head output
Args: Args:
input (list): List of Variables, output of backbone stages input (list): List of Variables, output of backbone stages
is_train (bool): whether in train or test mode is_train (bool): whether in train or test mode
Returns: Returns:
outputs (list): Variables of each output layer outputs (list): Variables of each output layer
""" """
...@@ -838,3 +837,232 @@ class PPYOLOTinyHead(YOLOv3Head): ...@@ -838,3 +837,232 @@ class PPYOLOTinyHead(YOLOv3Head):
route = self._upsample(route) route = self._upsample(route)
return outputs return outputs
@register
class YOLOv3PANHead(YOLOv3Head):
"""
Head block for YOLOv3PANHead network
Args:
conv_block_num (int): number of conv block in each detection block
norm_decay (float): weight decay for normalization layer weights
num_classes (int): number of output classes
anchors (list): anchors
anchor_masks (list): anchor masks
nms (object): an instance of `MultiClassNMS`
"""
__inject__ = ['yolo_loss', 'nms']
__shared__ = ['num_classes', 'weight_prefix_name']
def __init__(self,
conv_block_num=3,
norm_decay=0.,
num_classes=80,
anchors=[[10, 13], [16, 30], [33, 23], [30, 61], [62, 45],
[59, 119], [116, 90], [156, 198], [373, 326]],
anchor_masks=[[6, 7, 8], [3, 4, 5], [0, 1, 2]],
drop_block=False,
iou_aware=False,
iou_aware_factor=0.4,
block_size=3,
keep_prob=0.9,
yolo_loss="YOLOv3Loss",
spp=False,
nms=MultiClassNMS(
score_threshold=0.01,
nms_top_k=1000,
keep_top_k=100,
nms_threshold=0.45,
background_label=-1).__dict__,
weight_prefix_name='',
downsample=[32, 16, 8],
scale_x_y=1.0,
clip_bbox=True,
act='mish'):
super(YOLOv3PANHead, self).__init__(
conv_block_num=conv_block_num,
norm_decay=norm_decay,
num_classes=num_classes,
anchors=anchors,
anchor_masks=anchor_masks,
drop_block=drop_block,
iou_aware=iou_aware,
iou_aware_factor=iou_aware_factor,
block_size=block_size,
keep_prob=keep_prob,
yolo_loss=yolo_loss,
spp=spp,
nms=nms,
weight_prefix_name=weight_prefix_name,
downsample=downsample,
scale_x_y=scale_x_y,
clip_bbox=clip_bbox)
self.act = act
def _detection_block(self,
input,
channel,
conv_block_num=2,
is_first=False,
is_test=True,
name=None):
conv_left = self._conv_bn(
input,
channel,
act=self.act,
filter_size=1,
stride=1,
padding=0,
name='{}.left'.format(name))
conv_right = self._conv_bn(
input,
channel,
act=self.act,
filter_size=1,
stride=1,
padding=0,
name='{}.right'.format(name))
for j in range(conv_block_num):
conv_left = self._conv_bn(
conv_left,
channel,
act=self.act,
filter_size=1,
stride=1,
padding=0,
name='{}.left.{}'.format(name, 2 * j))
if self.use_spp and is_first and j == 1:
c = conv_left.shape[1]
conv_left = self._spp_module(conv_left, name="spp")
conv_left = self._conv_bn(
conv_left,
c,
act=self.act,
filter_size=1,
stride=1,
padding=0,
name='{}.left.{}'.format(name, 2 * j + 1))
else:
conv_left = self._conv_bn(
conv_left,
channel,
act=self.act,
filter_size=3,
stride=1,
padding=1,
name='{}.left.{}'.format(name, 2 * j + 1))
if self.drop_block and j == 1:
conv_left = DropBlock(
conv_left,
block_size=self.block_size,
keep_prob=self.keep_prob,
is_test=is_test)
conv = fluid.layers.concat(input=[conv_left, conv_right], axis=1)
conv = self._conv_bn(
conv,
channel * 2,
act=self.act,
filter_size=1,
stride=1,
padding=0,
name=name)
return conv, conv
def _get_outputs(self, input, is_train=True):
"""
Get YOLOv3 head output
Args:
input (list): List of Variables, output of backbone stages
is_train (bool): whether in train or test mode
Returns:
outputs (list): Variables of each output layer
"""
# get last out_layer_num blocks in reverse order
out_layer_num = len(self.anchor_masks)
blocks = input[-1:-out_layer_num - 1:-1]
# fpn
yolo_feats = []
route = None
for i, block in enumerate(blocks):
if i > 0: # perform concat in first 2 detection_block
block = fluid.layers.concat(input=[route, block], axis=1)
route, tip = self._detection_block(
block,
channel=512 // (2**i),
is_first=i == 0,
is_test=(not is_train),
conv_block_num=self.conv_block_num,
name=self.prefix_name + "fpn.{}".format(i))
yolo_feats.append(tip)
if i < len(blocks) - 1:
# do not perform upsample in the last detection_block
route = self._conv_bn(
input=route,
ch_out=512 // (2**i),
filter_size=1,
stride=1,
padding=0,
act=self.act,
name=self.prefix_name + "fpn_transition.{}".format(i))
# upsample
route = self._upsample(route)
# pan
pan_feats = [yolo_feats[-1]]
route = yolo_feats[out_layer_num - 1]
for i in reversed(range(out_layer_num - 1)):
channel = 512 // (2**i)
route = self._conv_bn(
input=route,
ch_out=channel,
filter_size=3,
stride=2,
padding=1,
act=self.act,
name=self.prefix_name + "pan_transition.{}".format(i))
block = yolo_feats[i]
block = fluid.layers.concat(input=[route, block], axis=1)
route, tip = self._detection_block(
block,
channel=channel,
is_first=False,
is_test=(not is_train),
conv_block_num=self.conv_block_num,
name=self.prefix_name + "pan.{}".format(i))
pan_feats.append(tip)
pan_feats = pan_feats[::-1]
outputs = []
for i, block in enumerate(pan_feats):
if self.iou_aware:
num_filters = len(self.anchor_masks[i]) * (self.num_classes + 6)
else:
num_filters = len(self.anchor_masks[i]) * (self.num_classes + 5)
with fluid.name_scope('yolo_output'):
block_out = fluid.layers.conv2d(
input=block,
num_filters=num_filters,
filter_size=1,
stride=1,
padding=0,
act=None,
param_attr=ParamAttr(
name=self.prefix_name +
"yolo_output.{}.conv.weights".format(i)),
bias_attr=ParamAttr(
regularizer=L2Decay(0.),
name=self.prefix_name +
"yolo_output.{}.conv.bias".format(i)))
outputs.append(block_out)
return outputs
...@@ -74,6 +74,7 @@ class IouAwareLoss(IouLoss): ...@@ -74,6 +74,7 @@ class IouAwareLoss(IouLoss):
iouk = self._iou(pred, gt, ioup, eps) iouk = self._iou(pred, gt, ioup, eps)
iouk.stop_gradient = True iouk.stop_gradient = True
loss_iou_aware = fluid.layers.cross_entropy(ioup, iouk, soft_label=True) loss_iou_aware = fluid.layers.sigmoid_cross_entropy_with_logits(ioup,
iouk)
loss_iou_aware = loss_iou_aware * self._loss_weight loss_iou_aware = loss_iou_aware * self._loss_weight
return loss_iou_aware return loss_iou_aware
...@@ -238,7 +238,6 @@ class YOLOv3Loss(object): ...@@ -238,7 +238,6 @@ class YOLOv3Loss(object):
along channel dimension along channel dimension
""" """
ioup = fluid.layers.slice(output, axes=[1], starts=[0], ends=[an_num]) ioup = fluid.layers.slice(output, axes=[1], starts=[0], ends=[an_num])
ioup = fluid.layers.sigmoid(ioup)
oriout = fluid.layers.slice( oriout = fluid.layers.slice(
output, output,
axes=[1], axes=[1],
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册