wdxpython
wdxpython
如果用1088，608的输入就没有问题，但是我都改成320了，为啥测试就不对呢，有哪些地方需要注意的吗？
可是我训练也用的320
可能我没有说清楚，如果用1088，608训练1088，608测试，没问题，测试正常。如果320，320训练，320，320测试，什么都检测不到，尺寸的地方我确定全部都改了，还有什么其他的参数需要修改吗？
May I ask the author what is the problem?
> It's all warnings, right? I got same but training and inference still normal Thank you for your answer, but I will crash in the middle of the training process....
> It's all warnings, right? I got same but training and inference still normal And this situation is random, once in 2 epoch and once in 10 epoch, which makes...
> Yeah I've encountered the same issue at the weighted focal loss, and it always happened around epoch 21. The dimension mismatch is always off by 1 right? I had...
> How much classes you train at same time? I occur same error in different framework (but still mmdet) when pass wrong class indexes (e.g index 15 when using only...
> > RuntimeError: CUDA error: device-side assert triggered > > terminate called after throwing an instance of 'c10::Error' > > Every time that I've gotten this error with this codebase,...
Hello author, this bug has troubled me for several days, how can I solve it?