Closed m-richa closed 11 months ago
Hi, usually the separation becomes fixed at quite early iterations, so 100k is definitely enough. May I ask what configuration did you use to train the model? Also please note that due to the strong view dependency of camera/hand in this dataset, it's sometimes quite difficult for the model to perfectly tell whether the shadow is a view-dependent effect or dynamic effect without any additional priors. Hence it is expected to have some artifacts in the separation.
Hope that helps!
After training for 100000 steps, the static part still has hand shadows. Does it need to be trained more?