sammilei

Hi, I encountered the same issue! Did you get it fixed?

> same here, any progress? No so I wrote my own code to convert the original segmenter model to onnx. That's fairly easy using the onnx documentation

My solution: ``` loss = loss.requires_grad_() # added thsi line because loss is not grad-able. loss.backward() ``` CUDA Version: 11.2

> @sefaburakokcu have you exported ONNX outputs with both the classtoken + patchtokens? Would be nice to have some guidance for this.

maybe try this: https://github.com/Nanne/pytorch-NetVlad