It seems that the model does not work for the provided input pairs of images and depths. about adampi HOT 2 CLOSED

@JinWonjoon

For the first question, you need to specify --disp_path when testing our method. By default, the --disp_path is set to the depth of 0810 (the squirrel example), and that's why the 3d photo of the penguin example seems strange.

For the second question, MPI cannot synthesize out-of-fov contents. The stretched pixels on the boundary are the out-of-fov regions; their color is determined by the padding mode of 'grid_sample', not synthesized by our network.

from adampi.

Oh, it was my mistakes!
Thanks for your help :)

from adampi.