The depth values stored in 16bit png files range from approximately 2k to 43k. What are these values? How can I interpret metric distance (mm) from these values? @anitarau

Created by Alwyn Mathew alwynmathew
The raw images are 16-bits which is the source of all these confusions. Converting to 8-bit fixes the issues.
Is this the correct way to read the depth files? @anitarau ``` depth = PIL.Image("/path/to/depth/file.png") depth = np.array(depth) # depth values in approx [2k, 43k] depth_norm = (depth - np.min(depth)) / (np.max(depth) - np.min(depth)) # normalized depth values in [0, 1] depth_acutal = depth_norm * 20 # depth values in [0, 20] ``` Normalization will not be correct if `np.max(depth)` and `np.min(depth)` varies from file to file.
@alwynmathew Yes, that's correct!
@anitarau Thank you. Does this same depth scaling apply to http://cmic.cs.ucl.ac.uk/ColonoscopyDepth/ too? Depth values correspond to [0,20] cm?
Thanks for your question! @alwynmathew the depth should be read as greyscale images in a [0,1] range. These values will correspond to [0,20] cm. The camera poses are also given in cm. @fjia the depth values do in fact correspond to a metric in the real world. The data is extracted from a computer tomography scan that preserves scale.
I think in colonoscopy - single camera, the depth is not perfect as 3D stereo reconstruction, they are depth value with scaling factors, so the distance maybe not in mm physical metric.

Interpert ground truth depth map page is loading…