input size

[yudawen]

I want to know the aspect ratio of an equirectangular image should be 1:2, why the author set the Omni-MNIST with a size of 60×60. For spherical convolution, can the setting of aspect ratio 1:1 work as well? @ChiWeiHsiao

[ChiWeiHsiao]

I am not sure whether using ratio 1:2 will result to better result than 1:1. You could try the 1:2 setting by specifying the outshape argument in line 119, 120 of example.py:

• OmniMNIST(fov=120, flip=True, h_rotate=True, v_rotate=True, train=True)
• -->OmniMNIST(fov=120, flip=True, h_rotate=True, v_rotate=True, train=True, outshape=(60, 120))

It will be great if you can share the result with ration 1:2.