search

JiaDingCN / GeminiFusion

MIT License
59 stars 7 forks source link

Doubt regarding the normalization of the depthmap during preprocessing of dataset in the training procedure #11

Open iacopo97 opened 2 months ago

iacopo97 commented 2 months ago

Thank you for your awesome work!!! I would like to be clarified of a doubt I have. During the normalization procedure are you dividing the values of the depthmap by 255? Is this the only normalization you applied to the depth in the dataset NYUDv2, even if it is a uint16 datum?

JiaDingCN commented 2 months ago

Hi, you may refer to the transform part. Does that solve your question about normalization procedure?

iacopo97 commented 2 months ago

Thank you very much!! I don't understand though why you divide the depthmap by 255. Is your depthmap in HHA format? From the dataset Nyudv2 that you provided seems to me that the depthmap is a uint16 matrix, therefore I would have divide the values of the map by 2**16 -1 not 255.

iacopo97 commented 2 months ago

I have another question about the evaluation on NYUDV2 dataset. Is it performed single scale or multiscale?