From f00395437d2eff5fd7fec8085bd1bbd80e8385ca Mon Sep 17 00:00:00 2001 From: Guanghua Yu <742925032@qq.com> Date: Tue, 17 Jan 2023 17:13:59 +0800 Subject: [PATCH] [cherry-pick] add YOLOv8 ACT demo (#7625) --- deploy/auto_compression/README.md | 57 +++++++++++++------ .../configs/picodet_s_qat_dis.yaml | 4 +- .../configs/ppyoloe_l_qat_dis.yaml | 5 +- .../configs/ppyoloe_plus_l_qat_dis.yaml | 5 +- .../configs/ppyoloe_plus_m_qat_dis.yaml | 5 +- .../configs/ppyoloe_plus_s_qat_dis.yaml | 5 +- .../configs/ppyoloe_plus_x_qat_dis.yaml | 5 +- .../configs/yolov5_s_qat_dis.yml | 5 +- .../configs/yolov6mt_s_qat_dis.yaml | 5 +- .../configs/yolov7_l_qat_dis.yaml | 5 +- .../configs/yolov8_reader.yml | 27 +++++++++ .../configs/yolov8_s_qat_dis.yaml | 32 +++++++++++ deploy/auto_compression/run.py | 14 ++++- requirements.txt | 2 +- 14 files changed, 128 insertions(+), 48 deletions(-) create mode 100644 deploy/auto_compression/configs/yolov8_reader.yml create mode 100644 deploy/auto_compression/configs/yolov8_s_qat_dis.yaml diff --git a/deploy/auto_compression/README.md b/deploy/auto_compression/README.md index 7b6657406..06c3c2f11 100644 --- a/deploy/auto_compression/README.md +++ b/deploy/auto_compression/README.md @@ -17,43 +17,52 @@ ## 2.Benchmark -### PP-YOLOE +### PP-YOLOE+ | 模型 | Base mAP | 离线量化mAP | ACT量化mAP | TRT-FP32 | TRT-FP16 | TRT-INT8 | 配置文件 | 量化模型 | | :-------- |:-------- |:--------: | :---------------------: | :----------------: | :----------------: | :---------------: | :----------------------: | :---------------------: | -| PP-YOLOE-l | 50.9 | - | 50.6 | 11.2ms | 7.7ms | **6.7ms** | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.5/deploy/auto_compression/configs/ppyoloe_l_qat_dis.yaml) | [Quant Model](https://bj.bcebos.com/v1/paddle-slim-models/act/ppyoloe_crn_l_300e_coco_quant.tar) | +| PP-YOLOE+_s | 43.7 | - | 42.9 | - | - | - | [config](./configs/ppyoloe_plus_s_qat_dis.yaml) | [Quant Model](https://bj.bcebos.com/v1/paddledet/deploy/Inference/ppyoloe_plus_s_qat_dis.tar) | +| PP-YOLOE+_m | 49.8 | - | 49.3 | - | - | - | [config](./configs/ppyoloe_plus_m_qat_dis.yaml) | [Quant Model](https://bj.bcebos.com/v1/paddledet/deploy/Inference/ppyoloe_plus_m_qat_dis.tar) | +| PP-YOLOE+_l | 52.9 | - | 52.6 | - | - | - | [config](./configs/ppyoloe_plus_l_qat_dis.yaml) | [Quant Model](https://bj.bcebos.com/v1/paddledet/deploy/Inference/ppyoloe_plus_l_qat_dis.tar) | +| PP-YOLOE+_x | 54.7 | - | 54.4 | - | - | - | [config](./configs/ppyoloe_plus_x_qat_dis.yaml) | [Quant Model](https://bj.bcebos.com/v1/paddledet/deploy/Inference/ppyoloe_plus_x_qat_dis.tar) | - mAP的指标均在COCO val2017数据集中评测得到，IoU=0.5:0.95。 -- PP-YOLOE-l模型在Tesla V100的GPU环境下测试，并且开启TensorRT，batch_size=1，包含NMS，测试脚本是[benchmark demo](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.4/deploy/python)。 -### PP-PicoDet +### YOLOv8 -| 模型 | 策略 | mAP | FP32 | FP16 | INT8 | 配置文件 | 模型 | -| :-------- |:-------- |:--------: | :----------------: | :----------------: | :---------------: | :----------------------: | :---------------------: | -| PicoDet-S-NPU | Baseline | 30.1 | - | - | - | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.5/configs/picodet/picodet_s_416_coco_npu.yml) | [Model](https://bj.bcebos.com/v1/paddle-slim-models/act/picodet_s_416_coco_npu.tar) | -| PicoDet-S-NPU | 量化训练 | 29.7 | - | - | - | [config](https://github.com/PaddlePaddle/PaddleSlim/tree/develop/demo/full_quantization/detection/configs/picodet_s_qat_dis.yaml) | [Model](https://bj.bcebos.com/v1/paddle-slim-models/act/picodet_s_npu_quant.tar) | +| 模型 | Base mAP | 离线量化mAP | ACT量化mAP | TRT-FP32 | TRT-FP16 | TRT-INT8 | 配置文件 | 量化模型 | +| :-------- |:-------- |:--------: | :---------------------: | :----------------: | :----------------: | :---------------: | :----------------------: | :---------------------: | +| YOLOv8-s | 44.9 | 43.9 | 44.3 | 9.27ms | 4.65ms | **3.78ms** | [config](./configs/yolov8_s_qat_dis.yaml) | [Model](https://bj.bcebos.com/v1/paddle-slim-models/act/yolov8_s_500e_coco_trt_nms_quant.tar) | +**注意：** +- 表格中YOLOv8模型均为带NMS的模型，可直接在TRT中部署，如果需要对齐测试标准，需要测试不带NMS的模型。 - mAP的指标均在COCO val2017数据集中评测得到，IoU=0.5:0.95。 +- 表格中的性能在Tesla T4的GPU环境下测试，并且开启TensorRT，batch_size=1。 -### PP-YOLOE+ +### PP-YOLOE | 模型 | Base mAP | 离线量化mAP | ACT量化mAP | TRT-FP32 | TRT-FP16 | TRT-INT8 | 配置文件 | 量化模型 | | :-------- |:-------- |:--------: | :---------------------: | :----------------: | :----------------: | :---------------: | :----------------------: | :---------------------: | -| PP-YOLOE+_s | 43.7 | - | 42.9 | - | - | - | [config](./configs/ppyoloe_plus_s_qat_dis.yaml) | [Quant Model](https://bj.bcebos.com/v1/paddledet/deploy/Inference/ppyoloe_plus_s_qat_dis.tar) | -| PP-YOLOE+_m | 49.8 | - | 49.3 | - | - | - | [config](./configs/ppyoloe_plus_m_qat_dis.yaml) | [Quant Model](https://bj.bcebos.com/v1/paddledet/deploy/Inference/ppyoloe_plus_m_qat_dis.tar) | -| PP-YOLOE+_l | 52.9 | - | 52.6 | - | - | - | [config](./configs/ppyoloe_plus_l_qat_dis.yaml) | [Quant Model](https://bj.bcebos.com/v1/paddledet/deploy/Inference/ppyoloe_plus_l_qat_dis.tar) | -| PP-YOLOE+_x | 54.7 | - | 54.4 | - | - | - | [config](./configs/ppyoloe_plus_x_qat_dis.yaml) | [Quant Model](https://bj.bcebos.com/v1/paddledet/deploy/Inference/ppyoloe_plus_x_qat_dis.tar) | +| PP-YOLOE-l | 50.9 | - | 50.6 | 11.2ms | 7.7ms | **6.7ms** | [config](./configs/ppyoloe_l_qat_dis.yaml) | [Quant Model](https://bj.bcebos.com/v1/paddle-slim-models/act/ppyoloe_crn_l_300e_coco_quant.tar) | - mAP的指标均在COCO val2017数据集中评测得到，IoU=0.5:0.95。 +- PP-YOLOE-l模型在Tesla V100的GPU环境下测试，并且开启TensorRT，batch_size=1，包含NMS，测试脚本是[benchmark demo](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.5/deploy/python)。 + +### PP-PicoDet +| 模型 | 策略 | mAP | FP32 | FP16 | INT8 | 配置文件 | 模型 | +| :-------- |:-------- |:--------: | :----------------: | :----------------: | :---------------: | :----------------------: | :---------------------: | +| PicoDet-S-NPU | Baseline | 30.1 | - | - | - | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.5/configs/picodet/picodet_s_416_coco_npu.yml) | [Model](https://bj.bcebos.com/v1/paddle-slim-models/act/picodet_s_416_coco_npu.tar) | +| PicoDet-S-NPU | 量化训练 | 29.7 | - | - | - | [config](https://github.com/PaddlePaddle/PaddleSlim/tree/develop/demo/full_quantization/detection/configs/picodet_s_qat_dis.yaml) | [Model](https://bj.bcebos.com/v1/paddle-slim-models/act/picodet_s_npu_quant.tar) | +- mAP的指标均在COCO val2017数据集中评测得到，IoU=0.5:0.95。 ## 3. 自动压缩流程 #### 3.1 准备环境 -- PaddlePaddle >= 2.3 （可从[Paddle官网](https://www.paddlepaddle.org.cn/install/quick?docurl=/documentation/docs/zh/install/pip/linux-pip.html)下载安装） -- PaddleSlim >= 2.3 -- PaddleDet >= 2.4 +- PaddlePaddle >= 2.4 （可从[Paddle官网](https://www.paddlepaddle.org.cn/install/quick?docurl=/documentation/docs/zh/install/pip/linux-pip.html)下载安装） +- PaddleSlim >= 2.4.1 +- PaddleDet >= 2.5 - opencv-python 安装paddlepaddle： @@ -74,9 +83,11 @@ pip install paddleslim pip install paddledet ``` +**注意：** YOLOv8模型的自动化压缩需要依赖安装最新[Develop Paddle](https://www.paddlepaddle.org.cn/install/quick?docurl=/documentation/docs/zh/develop/install/pip/linux-pip.html)和[Develop PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim#%E5%AE%89%E8%A3%85)版本。 + #### 3.2 准备数据集 -本案例默认以COCO数据进行自动压缩实验，如果自定义COCO数据，或者其他格式数据，请参考[数据准备文档](https://github.com/PaddlePaddle/PaddleDetection/blob/release/2.4/docs/tutorials/PrepareDataSet.md) 来准备数据。 +本案例默认以COCO数据进行自动压缩实验，如果自定义COCO数据，或者其他格式数据，请参考[数据准备文档](https://github.com/PaddlePaddle/PaddleDetection/blob/release/2.5/docs/tutorials/PrepareDataSet.md) 来准备数据。 如果数据集为非COCO格式数据，请修改[configs](./configs)中reader配置文件中的Dataset字段。 @@ -102,6 +113,16 @@ python tools/export_model.py \ trt=True \ ``` +YOLOv8-s模型，包含NMS，具体可参考[YOLOv8模型文档](https://github.com/PaddlePaddle/PaddleYOLO/tree/release/2.5/configs/yolov8), 然后执行： +```shell +python tools/export_model.py \ + -c configs/yolov8/yolov8_s_500e_coco.yml \ + -o weights=https://paddledet.bj.bcebos.com/models/yolov8_s_500e_coco.pdparams \ + trt=True +``` + +如快速体验，可直接下载[YOLOv8-s导出模型](https://bj.bcebos.com/v1/paddle-slim-models/act/yolov8_s_500e_coco_trt_nms.tar) + #### 3.4 自动压缩并产出模型 蒸馏量化自动压缩示例通过run.py脚本启动，会使用接口```paddleslim.auto_compression.AutoCompression```对模型进行自动压缩。配置config文件中模型路径、蒸馏、量化、和训练等部分的参数，配置完成后便可对模型进行量化和蒸馏。具体运行命令为： @@ -131,4 +152,4 @@ python eval.py --config_path=./configs/ppyoloe_l_qat_dis.yaml ## 4.预测部署 -- 可以参考[PaddleDetection部署教程](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.4/deploy)，GPU上量化模型开启TensorRT并设置trt_int8模式进行部署。 +- 可以参考[PaddleDetection部署教程](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.5/deploy)，GPU上量化模型开启TensorRT并设置trt_int8模式进行部署。 diff --git a/deploy/auto_compression/configs/picodet_s_qat_dis.yaml b/deploy/auto_compression/configs/picodet_s_qat_dis.yaml index 8d9091569..a5012be15 100644 --- a/deploy/auto_compression/configs/picodet_s_qat_dis.yaml +++ b/deploy/auto_compression/configs/picodet_s_qat_dis.yaml @@ -1,6 +1,6 @@ Global: reader_config: ./configs/picodet_reader.yml - input_list: ['image', 'scale_factor'] + include_nms: True Evaluation: True model_dir: ./picodet_s_416_coco_npu/ model_filename: model.pdmodel @@ -10,7 +10,7 @@ Distillation: alpha: 1.0 loss: l2 -Quantization: +QuantAware: use_pact: true activation_quantize_type: 'moving_average_abs_max' weight_bits: 8 diff --git a/deploy/auto_compression/configs/ppyoloe_l_qat_dis.yaml b/deploy/auto_compression/configs/ppyoloe_l_qat_dis.yaml index 4c7b684ea..df346d2b0 100644 --- a/deploy/auto_compression/configs/ppyoloe_l_qat_dis.yaml +++ b/deploy/auto_compression/configs/ppyoloe_l_qat_dis.yaml @@ -1,8 +1,7 @@ Global: reader_config: configs/ppyoloe_reader.yml - input_list: ['image', 'scale_factor'] - arch: YOLO + include_nms: True Evaluation: True model_dir: ./ppyoloe_crn_l_300e_coco model_filename: model.pdmodel @@ -12,7 +11,7 @@ Distillation: alpha: 1.0 loss: soft_label -Quantization: +QuantAware: use_pact: true activation_quantize_type: 'moving_average_abs_max' quantize_op_types: diff --git a/deploy/auto_compression/configs/ppyoloe_plus_l_qat_dis.yaml b/deploy/auto_compression/configs/ppyoloe_plus_l_qat_dis.yaml index 3e1b0ffc9..fd03aed09 100644 --- a/deploy/auto_compression/configs/ppyoloe_plus_l_qat_dis.yaml +++ b/deploy/auto_compression/configs/ppyoloe_plus_l_qat_dis.yaml @@ -1,8 +1,7 @@ Global: reader_config: configs/ppyoloe_plus_reader.yml - input_list: ['image', 'scale_factor'] - arch: YOLO + include_nms: True Evaluation: True model_dir: ./ppyoloe_plus_crn_l_80e_coco model_filename: model.pdmodel @@ -12,7 +11,7 @@ Distillation: alpha: 1.0 loss: soft_label -Quantization: +QuantAware: use_pact: true activation_quantize_type: 'moving_average_abs_max' quantize_op_types: diff --git a/deploy/auto_compression/configs/ppyoloe_plus_m_qat_dis.yaml b/deploy/auto_compression/configs/ppyoloe_plus_m_qat_dis.yaml index 26ee8fe7f..1da7fdb4b 100644 --- a/deploy/auto_compression/configs/ppyoloe_plus_m_qat_dis.yaml +++ b/deploy/auto_compression/configs/ppyoloe_plus_m_qat_dis.yaml @@ -1,8 +1,7 @@ Global: reader_config: configs/ppyoloe_plus_reader.yml - input_list: ['image', 'scale_factor'] - arch: YOLO + include_nms: True Evaluation: True model_dir: ./ppyoloe_plus_crn_m_80e_coco model_filename: model.pdmodel @@ -12,7 +11,7 @@ Distillation: alpha: 1.0 loss: soft_label -Quantization: +QuantAware: use_pact: true onnx_format: True activation_quantize_type: 'moving_average_abs_max' diff --git a/deploy/auto_compression/configs/ppyoloe_plus_s_qat_dis.yaml b/deploy/auto_compression/configs/ppyoloe_plus_s_qat_dis.yaml index c94350553..41bfde1e4 100644 --- a/deploy/auto_compression/configs/ppyoloe_plus_s_qat_dis.yaml +++ b/deploy/auto_compression/configs/ppyoloe_plus_s_qat_dis.yaml @@ -1,8 +1,7 @@ Global: reader_config: configs/ppyoloe_plus_reader.yml - input_list: ['image', 'scale_factor'] - arch: YOLO + include_nms: True Evaluation: True model_dir: ./ppyoloe_plus_crn_s_80e_coco model_filename: model.pdmodel @@ -12,7 +11,7 @@ Distillation: alpha: 1.0 loss: soft_label -Quantization: +QuantAware: use_pact: true activation_quantize_type: 'moving_average_abs_max' quantize_op_types: diff --git a/deploy/auto_compression/configs/ppyoloe_plus_x_qat_dis.yaml b/deploy/auto_compression/configs/ppyoloe_plus_x_qat_dis.yaml index 5cbed405c..4375a3844 100644 --- a/deploy/auto_compression/configs/ppyoloe_plus_x_qat_dis.yaml +++ b/deploy/auto_compression/configs/ppyoloe_plus_x_qat_dis.yaml @@ -1,8 +1,7 @@ Global: reader_config: configs/ppyoloe_plus_reader.yml - input_list: ['image', 'scale_factor'] - arch: YOLO + include_nms: True Evaluation: True model_dir: ./ppyoloe_plus_crn_x_80e_coco model_filename: model.pdmodel @@ -12,7 +11,7 @@ Distillation: alpha: 1.0 loss: soft_label -Quantization: +QuantAware: use_pact: true onnx_format: True activation_quantize_type: 'moving_average_abs_max' diff --git a/deploy/auto_compression/configs/yolov5_s_qat_dis.yml b/deploy/auto_compression/configs/yolov5_s_qat_dis.yml index 7111e27d4..309977ef6 100644 --- a/deploy/auto_compression/configs/yolov5_s_qat_dis.yml +++ b/deploy/auto_compression/configs/yolov5_s_qat_dis.yml @@ -1,8 +1,7 @@ Global: reader_config: configs/yolov5_reader.yml - input_list: ['image', 'scale_factor'] - arch: YOLO + include_nms: True Evaluation: True model_dir: ./yolov5_s_300e_coco model_filename: model.pdmodel @@ -12,7 +11,7 @@ Distillation: alpha: 1.0 loss: soft_label -Quantization: +QuantAware: use_pact: true activation_quantize_type: 'moving_average_abs_max' quantize_op_types: diff --git a/deploy/auto_compression/configs/yolov6mt_s_qat_dis.yaml b/deploy/auto_compression/configs/yolov6mt_s_qat_dis.yaml index 35a86ad21..e134494fe 100644 --- a/deploy/auto_compression/configs/yolov6mt_s_qat_dis.yaml +++ b/deploy/auto_compression/configs/yolov6mt_s_qat_dis.yaml @@ -1,8 +1,7 @@ Global: reader_config: configs/yolov5_reader.yml - input_list: ['image', 'scale_factor'] - arch: YOLO + include_nms: True Evaluation: True model_dir: ./yolov6mt_s_400e_coco model_filename: model.pdmodel @@ -12,7 +11,7 @@ Distillation: alpha: 1.0 loss: soft_label -Quantization: +QuantAware: activation_quantize_type: 'moving_average_abs_max' quantize_op_types: - conv2d diff --git a/deploy/auto_compression/configs/yolov7_l_qat_dis.yaml b/deploy/auto_compression/configs/yolov7_l_qat_dis.yaml index d233d6e76..801ccb405 100644 --- a/deploy/auto_compression/configs/yolov7_l_qat_dis.yaml +++ b/deploy/auto_compression/configs/yolov7_l_qat_dis.yaml @@ -1,8 +1,7 @@ Global: reader_config: configs/yolov5_reader.yml - input_list: ['image', 'scale_factor'] - arch: YOLO + include_nms: True Evaluation: True model_dir: ./yolov7_l_300e_coco model_filename: model.pdmodel @@ -12,7 +11,7 @@ Distillation: alpha: 1.0 loss: soft_label -Quantization: +QuantAware: activation_quantize_type: 'moving_average_abs_max' quantize_op_types: - conv2d diff --git a/deploy/auto_compression/configs/yolov8_reader.yml b/deploy/auto_compression/configs/yolov8_reader.yml new file mode 100644 index 000000000..202a49415 --- /dev/null +++ b/deploy/auto_compression/configs/yolov8_reader.yml @@ -0,0 +1,27 @@ +metric: COCO +num_classes: 80 + +# Dataset configuration +TrainDataset: + !COCODataSet + image_dir: train2017 + anno_path: annotations/instances_train2017.json + dataset_dir: dataset/coco/ + +EvalDataset: + !COCODataSet + image_dir: val2017 + anno_path: annotations/instances_val2017.json + dataset_dir: dataset/coco/ + +worker_num: 0 + +# preprocess reader in test +EvalReader: + sample_transforms: + - Decode: {} + - Resize: {target_size: [640, 640], keep_ratio: True, interp: 1} + - Pad: {size: [640, 640], fill_value: [114., 114., 114.]} + - NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none} + - Permute: {} + batch_size: 4 diff --git a/deploy/auto_compression/configs/yolov8_s_qat_dis.yaml b/deploy/auto_compression/configs/yolov8_s_qat_dis.yaml new file mode 100644 index 000000000..8c93e203e --- /dev/null +++ b/deploy/auto_compression/configs/yolov8_s_qat_dis.yaml @@ -0,0 +1,32 @@ + +Global: + reader_config: configs/yolov8_reader.yml + include_nms: True + Evaluation: True + model_dir: ./yolov8_s_500e_coco_trt_nms/ + model_filename: model.pdmodel + params_filename: model.pdiparams + +Distillation: + alpha: 1.0 + loss: soft_label + +QuantAware: + onnx_format: true + activation_quantize_type: 'moving_average_abs_max' + quantize_op_types: + - conv2d + - depthwise_conv2d + +TrainConfig: + train_iter: 8000 + eval_iter: 1000 + learning_rate: + type: CosineAnnealingDecay + learning_rate: 0.00003 + T_max: 10000 + optimizer_builder: + optimizer: + type: SGD + weight_decay: 4.0e-05 + diff --git a/deploy/auto_compression/run.py b/deploy/auto_compression/run.py index fe485fb8f..d940307db 100644 --- a/deploy/auto_compression/run.py +++ b/deploy/auto_compression/run.py @@ -23,6 +23,7 @@ from ppdet.metrics import COCOMetric, VOCMetric, KeyPointTopDownCOCOEval from paddleslim.auto_compression.config_helpers import load_config as load_slim_config from paddleslim.auto_compression import AutoCompression from post_process import PPYOLOEPostProcess +from paddleslim.common.dataloader import get_feed_vars def argsparser(): @@ -94,9 +95,12 @@ def eval_function(exe, compiled_test_program, test_feed_names, test_fetch_list): fetch_list=test_fetch_list, return_numpy=False) res = {} - if 'arch' in global_config and global_config['arch'] == 'PPYOLOE': - postprocess = PPYOLOEPostProcess( - score_threshold=0.01, nms_threshold=0.6) + if 'include_nms' in global_config and not global_config['include_nms']: + if 'arch' in global_config and global_config['arch'] == 'PPYOLOE': + postprocess = PPYOLOEPostProcess( + score_threshold=0.01, nms_threshold=0.6) + else: + assert "Not support arch={} now.".format(global_config['arch']) res = postprocess(np.array(outs[0]), data_all['scale_factor']) else: for out in outs: @@ -128,6 +132,10 @@ def main(): train_loader = create('EvalReader')(reader_cfg['TrainDataset'], reader_cfg['worker_num'], return_list=True) + if global_config.get('input_list') is None: + global_config['input_list'] = get_feed_vars( + global_config['model_dir'], global_config['model_filename'], + global_config['params_filename']) train_loader = reader_wrapper(train_loader, global_config['input_list']) if 'Evaluation' in global_config.keys() and global_config[ diff --git a/requirements.txt b/requirements.txt index 266d95ede..e789157f2 100644 --- a/requirements.txt +++ b/requirements.txt @@ -1,7 +1,7 @@ tqdm typeguard visualdl>=2.2.0 -opencv-python +opencv-python <= 4.6.0 PyYAML shapely scipy -- GitLab