Closed dhorka closed 2 years ago
Hi @dhorka , there are parameters I tuned to get better results. They don't have particular technical justification. You can see them as a post filtering on the predicted weights, where we discard small numbers.
The reason I use different thresholds is also for the performance. The above one (0.5) achieves better numerical performance in the whole testing set. The below one (0.35) seems to get better visual results on the provided examples.
Hi, I have a doubt related to the post-process done using the predicted values of the skinning weights. In the file run_skinning I found this line: https://github.com/zhan-xu/RigNet/blob/c5f5c985ca48700ac416b2f7f76a1fb3dde97554/run_skinning.py#L207
My doubt is why do you do that? I mean if I properly understand this line you are putting to zero all the values that accomplish this condition max_value/2. Is there any technical justification for it?
A related doubt is that I found that this line is different in the quick start script https://github.com/zhan-xu/RigNet/blob/c5f5c985ca48700ac416b2f7f76a1fb3dde97554/quick_start.py#L338 Is there any reason that instead of using 0.5 here you are using 0.35?