ssssober / depth-estimation Goto Github PK

The idea to predict mono depth with stereo networks is so AWESOME. However to my knowledge, disparities in active mono scene can be negative, given a reference plane. Perhaps it is better to take this into considerations?
Moreover, the dataset of speckled images is not available currently. I wonder on which synthesized stereo dataset did you use to apply speckles and how you apply them. Or, is there any plan of updating the Make-Your-Active-Data page in the future?

The checkpoint_20_0174000.tar file is corrupted. Can you upload a new file?

The pointclond looks good,looking forward to your page:Make-Your-Active-Data.