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.
994 stars 200 forks source link

MQBENCH 量化 报错 AttributeError: 'GraphModule' object has no attribute 'module' #83

Open Xiyuqiao opened 2 years ago

Xiyuqiao commented 2 years ago

训练是可以正常训练的，验证的时候就报错，大家有知道这是什么原因吗？

liu202204 commented 2 years ago

我也是这个问题，兄弟你是怎么解决的啊？ @Xiyuqiao

Xiyuqiao commented 2 years ago

我也不是很懂，暂时没有找到解决办法，先放弃了

liu202204 commented 2 years ago

我也不是很懂，暂时没有找到解决办法，先放弃了

你后面还有尝试其他人的量化代码嘛？

Xiyuqiao commented 2 years ago

我纯新手，所以不太懂哈，还找到了这两个，理论上肯定是可行的，但是还是会遇到一些报错，就放弃了，你可以试试看 https://github.com/maggiez0138/yolov5_quant_sample https://www.cnblogs.com/KdeS/p/15119831.html

zmzds commented 2 years ago

在插入量化节点之后，使用model.names = names,可以解决