Right now, the save_renders_pytorch3d function isn't particularly robust or flexible. Specifically, on this line, it casts the data to np.uint8. This may not be suitable for some applications, such as rendering a height map.
With float data, you wouldn't be able to do the compositing that we currently do, since it assumes the data can be visualized as a three-channel uint8. So handling this case would require some additional thinking.
Right now, the
save_renders_pytorch3dfunction isn't particularly robust or flexible. Specifically, on this line, it casts the data tonp.uint8. This may not be suitable for some applications, such as rendering a height map.With float data, you wouldn't be able to do the compositing that we currently do, since it assumes the data can be visualized as a three-channel
uint8. So handling this case would require some additional thinking.