MagicSource

@ayoolaolafenwa please try it again, it should supported now.

@wangyidong3 If the output says onnx saved into file, then ignore the last error. It was normal since I didn't fully catch the logic verification.

You can trace to onnx and deploy via some forward tools such like ncnn or tensorrt.

if the model params can be loaded it can be transferred. You can try replace the backbone pretrained weights. But if your backbone is not d2, you might need do...

I don't think this is not fit. It actually learned something. You need carefully check your lr, gama, steps, and even change the optimizer suite for your tiny dataset.

Can u try find out which op using inplace? I didn't get this before.

@acai66 Can u make a PR to solov2 if this works?

@sdimantsd Can u please pull and try again? it should be fixed.

@sdimantsd You are right, these files is for debugging GT is right or not. Can u verify is the GT is right or not from dataloader? You can send me...

@sdimantsd Can u please provide your machine config? I think this is gradient bloom. Clearly it won't learn anything.