Closed zjrandom951 closed 3 months ago
Was an image transposed fromportrait to landscape? It seems the intrinsics matrices rows are simply permuted as expected when transposing from portrait to landscape as in https://github.com/naver/dust3r/blob/main/dust3r/datasets/base/base_stereo_view_dataset.py#L202
when i try to run the training script, if the image resulotion is not square, then there is a probability that the diagonal element of the camera's intrinsics matrix is 0
thanks, it should be because the image was transposed.
There seems to be an issue with the camera's intrinsics, some focal is 0.