Closed Shaw-Way closed 8 months ago
When applying the sgt_loss to other monocular depth estimation models, did you also make the same modifications, such as not dividing the total loss by num_scale and dividing reprojection_loss by 2**scale?
Hello @Shaw-Way , Thanks for your interest!
I remember the settings you mentioned were following [1], and there are no special considerations when I set these parameters.
[1]. Hyunyoung Jung, Eunhyeok Park, and Sungjoo Yoo. Fine- grained semantics-aware representation enhancement for self-supervised monocular depth estimation. In Proceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 12642–12652, 2021.
ok, thanks for your reply!
Hi, author. Thanks for your remarkable work. https://github.com/xingyuuchen/tri-depth/blob/4ae4085caff74f5e62eb0377266b4a817f40cbee/manydepth/trainer.py#L667-L681 I noticed that the total_loss isn't divided by num_scales like original Manydepth does. Would this have an influence on the results? And I don't understand why smooth_loss and reprojection_loss are divided by 2**scale. https://github.com/xingyuuchen/tri-depth/blob/4ae4085caff74f5e62eb0377266b4a817f40cbee/manydepth/trainer.py#L667 https://github.com/xingyuuchen/tri-depth/blob/4ae4085caff74f5e62eb0377266b4a817f40cbee/manydepth/trainer.py#L673