BC #02: MMOneSphere, RGB Baseline with CNN
Hopefully flow does better. Actually it seems like it does? See the comparisons here
Created on May 3|Last edited on May 3
Comment
See the top plot which shows binary success rates with the naive RGB CNN method (regressing to 3DoF) and compare with PN++ with pointwise loss with flow. Do I have to tune the learning rate?
Results
Naive RGB
3
SVD Pointwise Flow
3
Example GIFs
(TODO...)
Add a comment