why need np.transpose(image,(2,0,1)) before feed image to model

[bigtree2020]

Hi I wonder why we need a np.transpose(image,(2,0,1)) before feed image to the model? https://github.com/intel-isl/MiDaS/blob/master/midas/transforms.py#L219

[dfrumkin]

@bigtree2020 It's a pytorch thing, unrelated to MiDaS; see, for example, https://discuss.pytorch.org/t/swap-axes-in-pytorch/970.

[AlexeyAB]

@bigtree2020

OpenCV img = cv2.imread(path) loads an image with HWC-layout (height, width, channels), while Pytorch requires CHW-layout. So we have to do np.transpose(image,(2,0,1)) for HWC->CHW transformation.

[mathmax12]

Thanks.

[Tar0kan]

@bigtree2020

OpenCV img = cv2.imread(path) loads an image with HWC-layout (height, width, channels), while Pytorch requires CHW-layout. So we have to do np.transpose(image,(2,0,1)) for HWC->CHW transformation.

Thank you, it helped me a lot!