1 源码
2 模型训练
数据集配置文件
训练命令
python tools/train.py -c configs/ppyolo/ppyolo_mbv3_large_coco.yml
加载预训练模型训练命令
python tools/train.py -c configs/picodet/application/mainbody_detection/picodet_lcnet_x2_5_640_mainbody.yml -o pretrain_weights=https://paddledet.bj.bcebos.com/models/picodet_lcnet_x2_5_640_mainbody.pdparams
模型预测
export CUDA_VISIBLE_DEVICES=0
python tools/infer.py -c configs/picodet/application/mainbody_detection/picodet_lcnet_x2_5_640_mainbody.yml --infer_img=your_image_path.jpg --output_dir=infer_output/ --draw_threshold=0.5 -o weights=output/picodet_lcnet_x2_5_640_mainbody/model_final
模型导出命令
python tools/export_model.py -c configs/ppyolo/ppyolo_mbv3_small_coco.yml --output_dir=./inference
导出某一预训练模型
python tools/export_model.py -c configs/picodet/application/mainbody_detection/picodet_lcnet_x2_5_640_mainbody.yml --output_dir=./inference -o weights=output/picodet_lcnet_x2_5_640_mainbody/model_final.pdparams
模型导出后有三个模型文件
3 模型C++部署
C++部署支持CPU以及GPU,具体部署流程:
参考这篇博文:paddle 目标检测C++部署流程