search

Gumpest / YOLOv5-Multibackbone-Compression

YOLOv5 Series Multi-backbone(TPH-YOLOv5, Ghostnet, ShuffleNetv2, Mobilenetv3Small, EfficientNetLite, PP-LCNet, SwinTransformer YOLO), Module(CBAM, DCN), Pruning (EagleEye, Network Slimming), Quantization (MQBench) and Deployment (TensorRT, ncnn) Compression Tool Box.
989 stars 201 forks source link

用大佬的正常训练的eagleeye剪枝方法的yaml跟用u神的正常训练出来权重文件有区别嘛 #56

Open suhanling opened 2 years ago

suhanling commented 2 years ago

如题，大佬，没有用大佬的正常训练的，是拿U神的原版v6版本训练出来的权重文件，拿来直接搜索最优子网然后微调不知道有没有影响，因为看到大佬的剪枝yaml里面有许多1.0的参数

ZhangHanN1 commented 2 years ago

我拿原版训练出来的参数文件进行eagleeye剪枝，得到的mAP一直是0