For training task model, what does `img_source img_traget lab_source lab_target` indicate respectively?

[lyndonzheng]

In our task, the img_source and lab_source indicate the synthetic image and depth respectively, and img_target and 'lab_target' indicate the real world image and depth respectively. If you want to do other tasks, you can change the label, like the semantic label.

[ksnzh]

If I use the task model and the L1 and smooth loss introduced in the article for supervised depth estimation, will the model achieve the reported score on NYU depth dataset as the f_T-only Benchmark Models shows?

[lyndonzheng]

Yes, for task model, you can change the img_source to NYU_TRAIN_IMAGE, lab_source to NYU_TRAIN_DEPTH, img_target to NYU_TEST_IMAGE, 'lab_target' to NYU_TEST_IMAGE.