Question about the output depth scale
Thanks for providing the implementation for this impressive work.
I have tried the demo on LUCES dataset by running eval_luces.py. The mean depth for the Ball object is 124 mm. The GT depth of this object is in the range of [0.94, 3.10] after we call NormalizeByMeanDepthLuces() in the luces_transforms, corresponding to the absolute depth range [117.5, 387.2] in mm.
I suppose after we get the output depth (corresponding to the variable output['depth_scales'][-1] in run_epoch.py), we should multiply it by the mean depth value to get the absolute depth map.
After I tried the above process, the outputs have reasonable relative shape, but the estimated depth is in the range of [51.5 mm, 139.02 mm]. May I ask whether my operation to get an absolute depth map is correct? If I miss something, please let me know.