@AdrianEddy Thank you for great works!
I try to inference with your code. but ,

Traceback (most recent call last):
File "main_stereo.py", line 612, in
main(args)
File "main_stereo.py", line 331, in main
inference_stereo(model_without_ddp,
File "anaconda3/envs/torch/lib/python3.8/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "unimatch/evaluate_stereo.py", line 799, in inference_stereo
pred_disp = model(left, right,
File "anaconda3/envs/torch/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "unimatch/unimatch/unimatch.py", line 190, in forward
feature0, feature1 = self.transformer(feature0, feature1,
File "anaconda3/envs/torch/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "unimatch/unimatch/transformer.py", line 272, in forward
shifted_window_attn_mask_1d = self.generate_shift_window_attn_mask_1d(
File "anaconda3/envs/torch/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "unimatch/unimatch/utils.py", line 207, in forward
mask_windows = window_partition_1d(img_mask, window_size_w) # nW, window_size, 1
File "unimatch/unimatch/utils.py", line 193, in window_partition_1d
B, W, C = x.shape
ValueError: too many values to unpack (expected 3)

@ylab604 Please check now

I see i did change the mask function when i check your code(yesterday).
But, important thing is that onnx graph(netron)is not normal campare with pinto0309

Anyway thank you for your kindness. And i will also update the result of excution

What do you mean it's not normal? What's weird about it?

What do you mean it's not normal? What's weird about it?

This means that if converted to onnx or jit, the inference output will be different from the original torch model.