BC03 3DoF, 1 Sphere, PN++ Per-Point Flow then Avg
The average is NON differentiable, just something we do after the fact.
Also unfortunately we can't look at the evaluation flow plots because those were not wrapped in the `utils.eval_mode()` and thus have the dropout making it look weird. :( Might want to re-do these, then.
Created on May 14|Last edited on May 17
Comment
Not sure why this doesn't do well. 0.079 among 5 trials.
The train/MSE_loss is also worse compared to PN++ with SVD, even though both are per-point on flow vectors in some sense (and mask out the tool). It's about 0.010 here versus 0.005 or earlier for PN++ SVD. Seems like the main issue again is sideways movement?
This is the older BC01 experiments we have with this setting, using ee2flow though:
Section 1
Run set
5
GIFs
Seems like again the main issue is just this sideways movement.
Add a comment