权重模型结构与core目录下代码模型结构对不上

[heiko134470-eng]

23-36-37权重模型结构与core目录下代码模型对不上，麻烦及时更新一下，谢谢！

[liyiersan]

I encountered the same issue. In the checkpoint, update_block.encoder.convc1 is defined as Conv2d(522, 56, kernel_size=(1, 1), stride=(1, 1)), whereas in the current model code it is Conv2d(522, 256, kernel_size=(1, 1), stride=(1, 1)).
I suspect this mismatch is the reason why the author chose to serialize the entire model object (via pickle) instead of saving a standard state_dict.

[liyiersan]

If you want to run Fast-FoundationStereo for inference on paired images, you can simply use the run_demo.py script. If you’d like to evaluate it on standard benchmarks, OpenStereo would be a good choice.