1. UserWarning: torch.meshgrid: in an upcoming release, it will be required to pass the indexing argument.
[Solution] Find the file in the directory D:\Users\JMan\anaconda3\envs\yolov5\Lib\site-packages\torch and make the following modificationsfunctional.py
# Remove this two weeks after landing.
kwargs = {
} if indexing is None else {
'indexing': indexing}
return _VF.meshgrid(tensors, **kwargs) # type: ignore[attr-defined]
Change to
# Remove this two weeks after landing.
kwargs = {
} if indexing is None else {
'indexing': indexing}
return _VF.meshgrid(tensors, **kwargs,indexing='ij') # type: ignore[attr-defined]
2. OSError: [WinError 1455] 页面文件太小，无法完成操作。
parser.add_argument('--workers', type=int, default=16, help='maximum number of dataloader workers')
The 16 in this line in train.py was changed to a small value. I changed it to 4 here and could barely run. That is to say, I can’t do porcelain work without diamonds...
3. OMP: Error #15: Initializing libiomp5md.dll, but found libiomp5md.dll already initialized.
The reason is that the torch package contains a file named libiomp5md.dll, which conflicts with the same file in the Anaconda environment, and the one in the Anaconda environment needs to be deleted.
Path D:\Users\JMan\anaconda3\envs\yolov5\Library\bin Find the file named libiomp5md.dll and delete it.
4.RuntimeError: result type Float can't be cast to the desired output type __int64
The reason is: During the loss calculation process, the numerical precision of Float appears, and a strong conversion of precision is required.
Solution: Modify two contents in [loss.py] in [utils].
Line 178 is modified as follows:
anchors, shape = self.anchors[i], p[i].shape
Line 211 is modified as follows:
indices.append((b, a, gj.clamp_(0, shape[2] - 1), gi.clamp_(0, shape[3] - 1))) # image, anchor, grid
training results
