-
Training loss goes to nan immediately when running texture prediction. Silhouette prediction succeeds with loss around 0.01. Predict textured (cell [38]) fails with loss nan. Is the notebook out of date? |
Beta Was this translation helpful? Give feedback.
Answered by
bottler
Jan 13, 2022
Replies: 1 comment 1 reply
-
This is a known issue, #561. Adding |
Beta Was this translation helpful? Give feedback.
1 reply
Answer selected by
lynnfyffe
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
This is a known issue, #561. Adding
perspective_correct=False
to everyRasterizationSettings
fixes it.