hotsuyuki/YOLOv5_PyTorch_cpp

模型导出

Opened this issue · 2 comments

JiangZongKang commented

请问在导出gpu版本的模型的时候，设置的model.model[-1].export = False吗？

hotsuyuki commented

Yes, I set model.model[-1].export = False when exporting TorchScript model (in both CPU and GPU).

Reference: https://github.com/yasenh/libtorch-yolov5#torchscript-model-export

JiangZongKang commented

Yes, I set model.model[-1].export = False when exporting TorchScript model (in both CPU and GPU).

Reference: https://github.com/yasenh/libtorch-yolov5#torchscript-model-export

But, when i set model.model[-1].export = False, this get TracerWarning: Converting a ce to be incorrect. We can't record the data flow of Python values, so this value will eans that the trace might not generalize to other inputs!
if self.grid[i].shape[2:4] != x[i].shape[2:4]:, does this affect libtorch load torchscript model?