Open yorklyb opened 6 months ago
Thank your for your question. The occlusion files are calculated with left right disparity consistency, though as stated, there are some sub folders, but using left right consistency, you can capture all. But anyway, since they are occlusion for stereo, when for monocular depth estimation, we did not use such files, refer to the dataloader and how we read data.
As for kitti, I just want to render some rgbs views, since the gT depth is sparse
Thanks for sharing the training scripts! I have two questions about the training data. In the data files, I noticed the use of occlusion files. Do they belong to the sub-dataset called Disparity Occlusion Weights? Also, for training on KITTI, it appears that only RGB images are used without depth supervision. I'm a bit confused about this.