Cannot converge with only RGB information

[pyni]

Hi~
i have tried to do pose optimization with your code.
i find a strange case that, if i use the mask of the rendered image (images[0][3, :,:]), it converges successfully.
But if i use RGB information of rendered image (images[0][0:3, :,:]), it fails.
I wonder whether the code can do back propagation with only RGB data of rendered image?

[Shichen Liu]

Hi there,

Backprop through RGB indeed might not as effective as the masks.

[pyni]

I see.
Thanks for quickly reply!