YOLO-人脸检测目标检测数据集-以PaddleDetection框架和YY-YOLOv2模型实现"人脸检测"+源代码+文档说明

# 压缩benchmark 在PaddleDetection, 提供了基于PaddleSlim进行模型压缩的完整教程和实验结果。详细教程请参考： - [量化](quantization) - [裁剪](prune) - [蒸馏](distillation) - [搜索](nas) 下面给出压缩的benchmark实验结果。 ## 测试环境 - Python 2.7.1 - PaddlePaddle >=1.6 - CUDA 9.0 - cuDNN >=7.4 - NCCL 2.1.2 ## 剪裁模型库 ### 训练策略 - 剪裁模型训练时使用[PaddleDetection模型库](https://paddledetection.readthedocs.io/MODEL_ZOO_cn.html)发布的模型权重作为预训练权重。 - 剪裁训练使用模型默认配置，即除`pretrained_weights`外配置不变。 - 剪裁模型全部为基于敏感度的卷积通道剪裁。 - YOLOv3模型主要剪裁`yolo_head`部分，即剪裁参数如下。 ``` --pruned_params="yolo_block.0.0.0.conv.weights,yolo_block.0.0.1.conv.weights,yolo_block.0.1.0.conv.weights,yolo_block.0.1.1.conv.weights,yolo_block.0.2.conv.weights,yolo_block.0.tip.conv.weights,yolo_block.1.0.0.conv.weights,yolo_block.1.0.1.conv.weights,yolo_block.1.1.0.conv.weights,yolo_block.1.1.1.conv.weights,yolo_block.1.2.conv.weights,yolo_block.1.tip.conv.weights,yolo_block.2.0.0.conv.weights,yolo_block.2.0.1.conv.weights,yolo_block.2.1.0.conv.weights,yolo_block.2.1.1.conv.weights,yolo_block.2.2.conv.weights,yolo_block.2.tip.conv.weights" ``` - YOLOv3模型剪裁中剪裁策略`r578`表示`yolo_head`中三个输出分支一次使用`0.5, 0.7, 0.8`的剪裁率剪裁，即剪裁率如下。 ``` --pruned_ratios="0.5,0.5,0.5,0.5,0.5,0.5,0.7,0.7,0.7,0.7,0.7,0.7,0.8,0.8,0.8,0.8,0.8,0.8" ``` - YOLOv3模型剪裁中剪裁策略`sensity`表示`yolo_head`中各参数剪裁率如下，该剪裁率为使用`yolov3_mobilnet_v1`模型在COCO数据集上敏感度实验分析得出。 ``` --pruned_ratios="0.1,0.2,0.2,0.2,0.2,0.1,0.2,0.3,0.3,0.3,0.2,0.1,0.3,0.4,0.4,0.4,0.4,0.3" ``` ### YOLOv3 on COCO | 骨架网络 | 剪裁策略 | GFLOPs | 模型体积(MB) | 输入尺寸 | Box AP | 下载 | | :----------------| :-------: | :------------: | :-------------: | :------: | :--------: | :-----------------------------------------------------: | | ResNet50-vd-dcn | baseline | 44.71 | 176.82 | 608 | 39.1 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/yolov3_r50vd_dcn.tar) | | ResNet50-vd-dcn | sensity | 37.53(-16.06%) | 149.49(-15.46%) | 608 | 39.8(+0.7) | [下载链接](https://paddlemodels.bj.bcebos.com/PaddleSlim/prune/yolov3_r50vd_dcn_prune1x.tar) | | ResNet50-vd-dcn | r578 | 29.98(-32.94%) | 112.08(-36.61%) | 608 | 38.3(-0.8) | [下载链接](https://paddlemodels.bj.bcebos.com/PaddleSlim/prune/yolov3_r50vd_dcn_prune578.tar) | | MobileNetV1 | baseline | 20.64 | 94.60 | 608 | 29.3 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/yolov3_mobilenet_v1.tar) | | MobileNetV1 | baseline | 9.66 | 94.60 | 416 | 29.3 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/yolov3_mobilenet_v1.tar) | | MobileNetV1 | baseline | 5.72 | 94.60 | 320 | 27.1 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/yolov3_mobilenet_v1.tar) | | MobileNetV1 | sensity | 13.57(-34.27%) | 67.60(-28.54%) | 608 | 30.2(+0.9) | [下载链接](https://paddlemodels.bj.bcebos.com/PaddleSlim/prune/yolov3_mobilenet_v1_prune1x.tar) | | MobileNetV1 | sensity | 6.35(-34.27%) | 67.60(-28.54%) | 416 | 29.7(+0.4) | [下载链接](https://paddlemodels.bj.bcebos.com/PaddleSlim/prune/yolov3_mobilenet_v1_prune1x.tar) | | MobileNetV1 | sensity | 3.76(-34.27%) | 67.60(-28.54%) | 320 | 27.2(+0.1) | [下载链接](https://paddlemodels.bj.bcebos.com/PaddleSlim/prune/yolov3_mobilenet_v1_prune1x.tar) | | MobileNetV1 | r578 | 6.27(-69.64%) | 31.30(-66.90%) | 608 | 27.8(-1.5) | [下载链接](https://paddlemodels.bj.bcebos.com/PaddleSlim/prune/yolov3_mobilenet_v1_prune578.tar) | | MobileNetV1 | r578 | 2.93(-69.64%) | 31.30(-66.90%) | 416 | 26.8(-2.5) | [下载链接](https://paddlemodels.bj.bcebos.com/PaddleSlim/prune/yolov3_mobilenet_v1_prune578.tar) | | MobileNetV1 | r578 | 1.74(-69.64%) | 31.30(-66.90%) | 320 | 24.0(-3.1) | [下载链接](https://paddlemodels.bj.bcebos.com/PaddleSlim/prune/yolov3_mobilenet_v1_prune578.tar) | - YOLO v3在训练阶段对minibatch采用随机reshape，可以采用相同的模型权重不同尺寸图片，表中`YOLOv3-MobileNetV1`提供了在`608/416/320`三种不同尺寸下的精度结果 - 在使用`sensity`剪裁策略下，`YOLOv3-ResNet50-vd-dcn`和`YOLOv3-MobileNetV1`分别减少了`16.06%`和`34.27%`的FLOPs，输入图像尺寸为608时精度分别提高`0.7`和`0.9` - 在使用`r578`剪裁策略下，`YOLOv3-ResNet50-vd-dcn`和`YOLOv3-MobileNetV1`分别减少了`32.98%`和`69.64%`的FLOPs，输入图像尺寸为608时精度分别降低`0.8`和`1.5` ### YOLOv3 on Pascal VOC | 骨架网络 | 剪裁策略 | GFLOPs | 模型体积(MB) | 输入尺寸 | Box AP | 下载 | | :----------------| :-------: | :------------: | :-------------: | :------: | :--------: | :-----------------------------------------------------: | | MobileNetV1 | baseline | 20.20 | 93.37 | 608 | 76.2 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/yolov3_mobilenet_v1_voc.tar) | | MobileNetV1 | baseline | 9.46 | 93.37 | 416 | 76.7 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/yolov3_mobilenet_v1_voc.tar) | | MobileNetV1 | baseline | 5.60 | 93.37 | 320 | 75.3 | [下载链接](https://paddlemodels.bj.bcebos.com/object_detection/yolov3_mobilenet_v1_voc.tar) | | MobileNetV1 | sensity | 13.22(-34.55%) | 66.53(-28.74%) | 608 | 78.4(+2.2) | [下载链接](https://paddlemodels.bj.bcebos.com/PaddleSlim/prune/yolov3_mobilenet_v1_voc_prune1x.tar) | | MobileNetV1 | sensity | 6.19(-34.55%) | 66.53(-28.74%) | 416 | 78.7(+2.0) | [下载链接](https://paddlemodels.bj.bcebos.com/PaddleSlim/prune/yolov3_mobilenet_v1_voc_prune1x.tar) | | MobileNetV1 | sensity | 3.66(-34.55%) | 66.53(-28.74%) | 320 | 76.1(+0.8) | [下载链接](https://paddlemodels.bj.bcebos.com/PaddleSlim/prune/yolov3_mobilenet_v1_voc_prune1x.tar) | | MobileNetV1 | r578 | 6.15(-69.57%) | 30.81(-67.00%) | 608 | 77.6(+1.4) | [下载链接](https://paddlemodels.bj.bcebos.com/PaddleSlim/prune/yolov3_mobilenet_v1_voc_prune578.tar) | | MobileNetV1 | r578 | 2.88(-69.57%) | 30.81(-67.00%) | 416 | 77.7(+1.0) | [下载链接](https://paddlemodels.bj.bcebos.com/PaddleSlim/prune/yolov3_mobilenet_v1_voc_prune578.tar) | | MobileNetV1 | r578 | 1.70(-69.57%) | 30.81(-67.00%) | 320 | 75.5(+0.2) | [下载链接](https://paddlemodels.bj.bcebos.com/PaddleSlim/prune/yolov3_mobilenet_v1_voc_prune578.tar) | - YOLO v3在训练阶段对minibatch采用随机reshape，可以采用相同的模型权重不同尺寸图片，表中`YOLOv3-MobileNetV1`提供了在`608/416/320`三种不同尺寸下的精度结果 - 在使用`sensity`和`r578`剪裁策略下，`YOLOv3-MobileNetV1`分别减少了`34.55%`和`69.57%`的FLOPs，输入图像尺寸为608时精度分别提高`2.2`和`1.4` ### 蒸馏通道剪裁模型 可通过高精度模型蒸馏通道剪裁后模型的方式，训练方法及相关示例见[蒸馏通道剪裁模型](https://github.com/PaddlePaddle/PaddleDetection/blob/develop/static/slim/extensions/distill_pruned_model/distill_pruned_model_demo.ipynb)。 COCO数据集上蒸馏通道剪裁模型库如下。 | 骨架网络 | 剪裁策略 | GFLOPs | 模型体积(MB) | 输入尺�