Hi, thanks for your great work about dataset and MidaS implementation.
I have two questions,
In the test_depth.py code, the depth_gt and depth_pred are both clamped to [0, 1] and then compute metrics like rmse. Why they should be clamped. How do I get the real metric depth to evaluate the performance on NYU_DEPTH_V2
The depth estimation results reported in the paper are a little strange because the L1 error and delta_1 are too low compared to the results reported in MidaS and DPT. Is there any difference in the metric computation between your repo and MidaS?
If there are any confused things I haven't clarified, please let me know! Hope for your reply, thx :)
Hi, thanks for your great work about dataset and MidaS implementation.
I have two questions,
If there are any confused things I haven't clarified, please let me know! Hope for your reply, thx :)