preprocess_image.py script does not work properly

[SrTartarus]

Description

Hello I tried to get my rgba, depth and normal images but I could not because the preprocess_image.py gives me a black image in the supposed normal image but the rgba image and depth if it gives me correctly I don't know what it could be because I tried some previous commits and the same thing happens, it's like if the pretrained model file doesn't works

To let you know that I tried with your images you have on ./data folder so I tried with the hamburger image but I've got same results.

UPDATE I used this commit c010e264725840051aec0a2ec05511de2a7831e7 before resize and recenter changes then I used the pretrained model v1 that uses UNet and that works I want to add that I tried omnidata demo from original repository that they have and still didn't work so I believe pretrained v2 for normals doesn't work

Steps to Reproduce

1.- Download pretrained models from gdown 2.- preprocess_image.py data/hamburger.png

Expected Behavior

Give the image of the normal map representing the object of your image

Environment

Google Colab

[ashawkey]

@SrTartarus Hi, normal seems OK on my side, could you check if the checkpoint is loading? Also, normal is actually not used currently, so you can modify the code to omit loading it.

[mochou-wujiu]

same problem too ... The normal png is black

[cwwjyh]

@SrTartarus Hi, normal seems OK on my side, could you check if the checkpoint is loading? Also, normal is actually not used currently, so you can modify the code to omit loading it.

I also meet this problem and how I solve it. I found that the problem might be caused by the pre-trained model. Looking forward to your reply! Thank you!