add SSLD model (#2421)

configs/face_detection/README.md

@@ -11,7 +11,7 @@
 
 | 网络结构 | 输入尺寸 | 图片个数/GPU | 学习率策略 | Easy/Medium/Hard Set  | 预测时延（SD855）| 模型大小(MB) | 下载 | 配置文件 |
 |:------------:|:--------:|:----:|:-------:|:-------:|:---------:|:----------:|:---------:|:--------:|
-| BlazeFace  | 640  |    8    | 1000e     | 0.889 / 0.859 / 0.740 | - | 0.472 |[下载链接](https://paddledet.bj.bcebos.com/models/blazeface_1000e.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/face_detection/blazeface_1000e.yml) |
+| BlazeFace  | 640  |    8    | 1000e     | 0.889 / 0.859 / 0.740 | - | 0.472 |[下载链接](https://paddledet.bj.bcebos.com/models/blazeface_1000e.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/face_detection/blazeface_1000e.yml) |
 
 **注意:**  
 - 我们使用多尺度评估策略得到`Easy/Medium/Hard Set`里的mAP。具体细节请参考[在WIDER-FACE数据集上评估](#在WIDER-FACE数据集上评估)。

configs/gn/README.md

@@ -4,10 +4,10 @@
 
 | 骨架网络         | 网络类型        | 每张GPU图片个数 | 学习率策略 |推理时间(fps)| Box AP | Mask AP |  下载  | 配置文件 |
 | :------------- | :------------- | :-----------: | :------: | :--------: |:-----: | :-----: | :----: | :----: |
-| ResNet50-FPN   | Faster         |    1          |   2x     |    -       |  41.9  |    -    | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_r50_fpn_gn_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/gn/faster_rcnn_r50_fpn_gn_2x_coco.yml) |
-| ResNet50-FPN   | Mask           |    1          |   2x     |    -       |  42.3   |  38.4  | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_r50_fpn_gn_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/gn/mask_rcnn_r50_fpn_gn_2x_coco.yml) |
-| ResNet50-FPN   | Cascade Faster    |    1          |   2x     |    -       |  -   |  -  | [下载链接](https://paddledet.bj.bcebos.com/models/cascade_rcnn_r50_fpn_gn_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/gn/cascade_rcnn_r50_fpn_gn_2x_coco.yml) |
-| ResNet50-FPN   | Cacade Mask      |    1          |   2x     |    -       |  45.0   |  39.3  | [下载链接](https://paddledet.bj.bcebos.com/models/cascade_mask_rcnn_r50_fpn_gn_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/gn/cascade_mask_rcnn_r50_fpn_gn_2x_coco.yml) |
+| ResNet50-FPN   | Faster         |    1          |   2x     |    -       |  41.9  |    -    | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_r50_fpn_gn_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/gn/faster_rcnn_r50_fpn_gn_2x_coco.yml) |
+| ResNet50-FPN   | Mask           |    1          |   2x     |    -       |  42.3   |  38.4  | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_r50_fpn_gn_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/gn/mask_rcnn_r50_fpn_gn_2x_coco.yml) |
+| ResNet50-FPN   | Cascade Faster    |    1          |   2x     |    -       |  44.6   |  -  | [下载链接](https://paddledet.bj.bcebos.com/models/cascade_rcnn_r50_fpn_gn_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/gn/cascade_rcnn_r50_fpn_gn_2x_coco.yml) |
+| ResNet50-FPN   | Cacade Mask      |    1          |   2x     |    -       |  45.0   |  39.3  | [下载链接](https://paddledet.bj.bcebos.com/models/cascade_mask_rcnn_r50_fpn_gn_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/gn/cascade_mask_rcnn_r50_fpn_gn_2x_coco.yml) |
 
 
 **注意：** Faster R-CNN baseline仅使用 `2fc` head，而此处使用[`4conv1fc` head](https://arxiv.org/abs/1803.08494)（4层conv之间使用GN），并且FPN也使用GN，而对于Mask R-CNN是在mask head的4层conv之间也使用GN。

configs/solov2/README.md

@@ -19,8 +19,8 @@ SOLOv2 (Segmenting Objects by Locations) is a fast instance segmentation framewo
 | BlendMask | R50-FPN | True        |   3x    |     37.8        |  13.5  | V100 |   -  |  -  |
 | SOLOv2 (Paper) | R50-FPN | False        |   1x    |     34.8        |  18.5  | V100 |   -  |  -  |
 | SOLOv2 (Paper) | X101-DCN-FPN | True        |   3x    |     42.4        |  5.9  | V100 |   -  |  -  |
-| SOLOv2 | R50-FPN                 |  False                |   1x    |    35.5         |  21.9     | V100 |  [model](https://paddledet.bj.bcebos.com/models/solov2_r50_fpn_1x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/solov2/solov2_r50_fpn_1x_coco.yml) |
-| SOLOv2 | R50-FPN                 |  True                |   3x    |     38.0         |   21.9    | V100 |  [model](https://paddledet.bj.bcebos.com/models/solov2_r50_fpn_3x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/dygraph/configs/solov2/solov2_r50_fpn_3x_coco.yml) |
+| SOLOv2 | R50-FPN                 |  False                |   1x    |    35.5         |  21.9     | V100 |  [model](https://paddledet.bj.bcebos.com/models/solov2_r50_fpn_1x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/solov2/solov2_r50_fpn_1x_coco.yml) |
+| SOLOv2 | R50-FPN                 |  True                |   3x    |     38.0         |   21.9    | V100 |  [model](https://paddledet.bj.bcebos.com/models/solov2_r50_fpn_3x_coco.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/solov2/solov2_r50_fpn_3x_coco.yml) |
 
 **Notes:**
 

configs/ssld/README.md

+### Simple semi-supervised label knowledge distillation solution (SSLD)
+
+## Model Zoo
+
+| 骨架网络              | 网络类型       | 每张GPU图片个数 | 学习率策略 |推理时间(fps) | Box AP | Mask AP |                           下载                          | 配置文件 |
+| :------------------- | :------------| :-----: | :-----: | :------------: | :-----: | :-----: | :-----------------------------------------------------: | :-----: |
+| ResNet50-vd-FPN      | Faster         |    1    |   1x    |     ----     |  41.4  |  -  | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_r50_vd_fpn_ssld_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ssld/faster_rcnn_r50_vd_fpn_ssld_1x_coco.yml) |
+| ResNet50-vd-FPN      | Faster         |    1    |   2x    |     ----     |  42.3  |  -  | [下载链接](https://paddledet.bj.bcebos.com/models/faster_rcnn_r50_vd_fpn_ssld_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ssld/faster_rcnn_r50_vd_ssld_fpn_2x_coco.yml) |
+| ResNet50-vd-FPN         | Mask         |    1    |   1x    |     ----     |  42.0  |    38.2   | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_r50_vd_fpn_ssld_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ssld/mask_rcnn_r50_vd_fpn_ssld_1x_coco.yml) |
+| ResNet50-vd-FPN         | Mask         |    1    |   2x    |     ----     |  42.7 |    38.9   | [下载链接](https://paddledet.bj.bcebos.com/models/mask_rcnn_r50_vd_fpn_ssld_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ssld/mask_rcnn_r50_vd_fpn_ssld_2x_coco.yml) |
+| ResNet50-vd-FPN         | Cascade Faster         |    1    |   1x    |     ----     |  44.4  |    -    | [下载链接](https://paddledet.bj.bcebos.com/models/cascade_rcnn_r50_vd_fpn_ssld_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ssld/cascade_rcnn_r50_vd_fpn_ssld_1x_coco.yml) |
+| ResNet50-vd-FPN         | Cascade Faster         |    1    |   2x    |     ----     |  45.0  |    -    | [下载链接](https://paddledet.bj.bcebos.com/models/cascade_rcnn_r50_vd_fpn_ssld_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ssld/cascade_rcnn_r50_vd_fpn_ssld_2x_coco.yml) |
+| ResNet50-vd-FPN         | Cascade Mask         |    1    |   1x    |     ----     |  44.9 |    39.1    | [下载链接](https://paddledet.bj.bcebos.com/models/cascade_mask_rcnn_r50_vd_fpn_ssld_1x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ssld/cascade_mask_rcnn_r50_vd_fpn_ssld_1x_coco.yml) |
+| ResNet50-vd-FPN         | Cascade Mask         |    1    |   2x    |     ----     |  45.7  |    39.7    | [下载链接](https://paddledet.bj.bcebos.com/models/cascade_mask_rcnn_r50_vd_fpn_ssld_2x_coco.pdparams) | [配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ssld/cascade_mask_rcnn_r50_vd_fpn_ssld_2x_coco.yml) |
+
+
+**注意事项:**  
+
+- [SSLD](https://arxiv.org/abs/1811.11168)是一种知识蒸馏方法，我们使用蒸馏后性能更强的backbone预训练模型，进一步提升检测精度，详细方案请参考[知识蒸馏教程](https://github.com/PaddlePaddle/PaddleClas/blob/develop/docs/en/advanced_tutorials/distillation/distillation_en.md)
+
+![demo image](../../docs/images/ssld_model.png)
+
+## Citations
+```
+@misc{cui2021selfsupervision,
+      title={Beyond Self-Supervision: A Simple Yet Effective Network Distillation Alternative to Improve Backbones},
+      author={Cheng Cui and Ruoyu Guo and Yuning Du and Dongliang He and Fu Li and Zewu Wu and Qiwen Liu and Shilei Wen and Jizhou Huang and Xiaoguang Hu and Dianhai Yu and Errui Ding and Yanjun Ma},
+      year={2021},
+      eprint={2103.05959},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV}
+}
+```

configs/ssld/cascade_mask_rcnn_r50_vd_fpn_ssld_1x_coco.yml

+_BASE_: [
+  '../datasets/coco_instance.yml',
+  '../runtime.yml',
+  '../cascade_rcnn/_base_/optimizer_1x.yml',
+  '../cascade_rcnn/_base_/cascade_mask_rcnn_r50_fpn.yml',
+  '../cascade_rcnn/_base_/cascade_mask_fpn_reader.yml',
+]
+pretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/ResNet50_vd_ssld_v2_pretrained.pdparams
+weights: output/cascade_mask_rcnn_r50_vd_fpn_ssld_1x_coco/model_final
+
+ResNet:
+  depth: 50
+  variant: d
+  norm_type: bn
+  freeze_at: 0
+  return_idx: [0,1,2,3]
+  num_stages: 4
+  lr_mult_list: [0.05, 0.05, 0.1, 0.15]

configs/ssld/cascade_mask_rcnn_r50_vd_fpn_ssld_2x_coco.yml

+_BASE_: [
+  '../datasets/coco_instance.yml',
+  '../runtime.yml',
+  '../cascade_rcnn/_base_/optimizer_1x.yml',
+  '../cascade_rcnn/_base_/cascade_mask_rcnn_r50_fpn.yml',
+  '../cascade_rcnn/_base_/cascade_mask_fpn_reader.yml',
+]
+pretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/ResNet50_vd_ssld_v2_pretrained.pdparams
+weights: output/cascade_mask_rcnn_r50_vd_fpn_ssld_2x_coco/model_final
+
+ResNet:
+  depth: 50
+  variant: d
+  norm_type: bn
+  freeze_at: 0
+  return_idx: [0,1,2,3]
+  num_stages: 4
+  lr_mult_list: [0.05, 0.05, 0.1, 0.15]
+
+epoch: 24
+LearningRate:
+  base_lr: 0.01
+  schedulers:
+  - !PiecewiseDecay
+    gamma: 0.1
+    milestones: [12, 22]
+  - !LinearWarmup
+    start_factor: 0.1
+    steps: 1000

configs/ssld/cascade_rcnn_r50_vd_fpn_ssld_1x_coco.yml

+_BASE_: [
+  '../datasets/coco_detection.yml',
+  '../runtime.yml',
+  '../cascade_rcnn/_base_/optimizer_1x.yml',
+  '../cascade_rcnn/_base_/cascade_rcnn_r50_fpn.yml',
+  '../cascade_rcnn/_base_/cascade_fpn_reader.yml',
+]
+pretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/ResNet50_vd_ssld_v2_pretrained.pdparams
+weights: output/cascade_rcnn_r50_vd_fpn_ssld_1x_coco/model_final
+
+ResNet:
+  depth: 50
+  variant: d
+  norm_type: bn
+  freeze_at: 0
+  return_idx: [0,1,2,3]
+  num_stages: 4
+  lr_mult_list: [0.05, 0.05, 0.1, 0.15]

configs/ssld/cascade_rcnn_r50_vd_fpn_ssld_2x_coco.yml

+_BASE_: [
+  '../datasets/coco_detection.yml',
+  '../runtime.yml',
+  '../cascade_rcnn/_base_/optimizer_1x.yml',
+  '../cascade_rcnn/_base_/cascade_rcnn_r50_fpn.yml',
+  '../cascade_rcnn/_base_/cascade_fpn_reader.yml',
+]
+pretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/ResNet50_vd_ssld_v2_pretrained.pdparams
+weights: output/cascade_rcnn_r50_vd_fpn_ssld_2x_coco/model_final
+
+ResNet:
+  depth: 50
+  variant: d
+  norm_type: bn
+  freeze_at: 0
+  return_idx: [0,1,2,3]
+  num_stages: 4
+  lr_mult_list: [0.05, 0.05, 0.1, 0.15]
+
+epoch: 24
+LearningRate:
+  base_lr: 0.01
+  schedulers:
+  - !PiecewiseDecay
+    gamma: 0.1
+    milestones: [12, 22]
+  - !LinearWarmup
+    start_factor: 0.1
+    steps: 1000

configs/ssld/faster_rcnn_r50_vd_fpn_ssld_1x_coco.yml

+_BASE_: [
+  '../datasets/coco_detection.yml',
+  '../runtime.yml',
+  '../faster_rcnn/_base_/optimizer_1x.yml',
+  '../faster_rcnn/_base_/faster_rcnn_r50_fpn.yml',
+  '../faster_rcnn/_base_/faster_fpn_reader.yml',
+]
+pretrain_weights:  https://paddledet.bj.bcebos.com/models/pretrained/ResNet50_vd_ssld_v2_pretrained.pdparams
+weights: output/faster_rcnn_r50_vd_fpn_ssld_1x_coco/model_final
+
+ResNet:
+  depth: 50
+  variant: d
+  norm_type: bn
+  freeze_at: 0
+  return_idx: [0,1,2,3]
+  num_stages: 4
+  lr_mult_list: [0.05, 0.05, 0.1, 0.15]
+
+epoch: 12
+LearningRate:
+  base_lr: 0.01
+  schedulers:
+  - !PiecewiseDecay
+    gamma: 0.1
+    milestones: [8, 11]
+  - !LinearWarmup
+    start_factor: 0.1
+    steps: 1000

configs/ssld/faster_rcnn_r50_vd_fpn_ssld_2x_coco.yml

+_BASE_: [
+  '../datasets/coco_detection.yml',
+  '../runtime.yml',
+  '../faster_rcnn/_base_/optimizer_1x.yml',
+  '../faster_rcnn/_base_/faster_rcnn_r50_fpn.yml',
+  '../faster_rcnn/_base_/faster_fpn_reader.yml',
+]
+pretrain_weights:  https://paddledet.bj.bcebos.com/models/pretrained/ResNet50_vd_ssld_v2_pretrained.pdparams
+weights: output/faster_rcnn_r50_vd_fpn_ssld_2x_coco/model_final
+
+ResNet:
+  depth: 50
+  variant: d
+  norm_type: bn
+  freeze_at: 0
+  return_idx: [0,1,2,3]
+  num_stages: 4
+  lr_mult_list: [0.05, 0.05, 0.1, 0.15]
+
+epoch: 24
+LearningRate:
+  base_lr: 0.01
+  schedulers:
+  - !PiecewiseDecay
+    gamma: 0.1
+    milestones: [12, 22]
+  - !LinearWarmup
+    start_factor: 0.1
+    steps: 1000

configs/ssld/mask_rcnn_r50_vd_fpn_ssld_1x_coco.yml

+_BASE_: [
+  '../datasets/coco_instance.yml',
+  '../runtime.yml',
+  '../mask_rcnn/_base_/optimizer_1x.yml',
+  '../mask_rcnn/_base_/mask_rcnn_r50_fpn.yml',
+  '../mask_rcnn/_base_/mask_fpn_reader.yml',
+]
+pretrain_weights:  https://paddledet.bj.bcebos.com/models/pretrained/ResNet50_vd_ssld_v2_pretrained.pdparams
+weights: output/mask_rcnn_r50_vd_fpn_ssld_1x_coco/model_final
+
+ResNet:
+  depth: 50
+  variant: d
+  norm_type: bn
+  freeze_at: 0
+  return_idx: [0,1,2,3]
+  num_stages: 4
+  lr_mult_list: [0.05, 0.05, 0.1, 0.15]
+
+epoch: 12
+LearningRate:
+  base_lr: 0.01
+  schedulers:
+  - !PiecewiseDecay
+    gamma: 0.1
+    milestones: [8, 11]
+  - !LinearWarmup
+    start_factor: 0.1
+    steps: 1000

configs/ssld/mask_rcnn_r50_vd_fpn_ssld_2x_coco.yml

+_BASE_: [
+  '../datasets/coco_instance.yml',
+  '../runtime.yml',
+  '../mask_rcnn/_base_/optimizer_1x.yml',
+  '../mask_rcnn/_base_/mask_rcnn_r50_fpn.yml',
+  '../mask_rcnn/_base_/mask_fpn_reader.yml',
+]
+pretrain_weights:  https://paddledet.bj.bcebos.com/models/pretrained/ResNet50_vd_ssld_v2_pretrained.pdparams
+weights: output/mask_rcnn_r50_vd_fpn_ssld_2x_coco/model_final
+
+ResNet:
+  depth: 50
+  variant: d
+  norm_type: bn
+  freeze_at: 0
+  return_idx: [0,1,2,3]
+  num_stages: 4
+  lr_mult_list: [0.05, 0.05, 0.1, 0.15]
+
+epoch: 24
+LearningRate:
+  base_lr: 0.01
+  schedulers:
+  - !PiecewiseDecay
+    gamma: 0.1
+    milestones: [12, 22]
+  - !LinearWarmup
+    start_factor: 0.1
+    steps: 1000