Closed Zarxrax closed 1 year ago
Added it just now. But i have no time to train it.
Thanks, I was able to train it for a few hours to test.
In train.py, I had to revert line 188 back to:
strategy="ddp_find_unused_parameters_false" if opt.devices > 1 else None,
Then the training worked.
I was not able to get export to onnx working. But I could test through the app.py. So far it does not seem to be as good as the other models, but I will try some experiments to get a better result.
Is it possible to add support for InSPyReNet to the training code? https://github.com/plemeri/inspyrenet
Their paper shows significant improvement from ISNet.