Different result for the same picture about depth2hha-python

[hyclover]

I tried to use RD = cv2.imread(os.path.join(root, '000001.png'), cv2.COLOR_BGR2GRAY) / 255 * max and just got the normal result. So the question is did I not use the right depth value before?

my final result

[charlesCXK]

@hyclover
Hi, I think the depth map should not be normalized or be viewed as a 3-channel color map. If you normalize the depth map, such as depths = depths / max * 255, the maximum depth value would be 255 meters, and the hyper-parameters used in our code will be not suitable.

[hyclover]

@hyclover
Hi, I think the depth map should not be normalized or be viewed as a 3-channel color map. If you normalize the depth map, such as depths = depths / max * 255, the maximum depth value would be 255 meters, and the hyper-parameters used in our code will be not suitable.

I just got the code from the Internet to extract the depth image. Would you mind sharing your code for extracting depth images from official Matlab file nyu_depth_v2_labeled.mat. Thanks very much!

[charlesCXK]

@hyclover
As you illustrated above, I think you just need to remove the normalization code. Finally, the depth value of NYU Depth v2 images will lie in [0, 20] meters for most cases, since it is an in-door dataset.