search

BeileiCui / EndoDAC

[MICCAI'2024] EndoDAC: Efficient Adapting Foundation Model for Self-Supervised Depth Estimation from Any Endoscopic Camera
30 stars 2 forks source link

Units for _disp.npy data #12

Open SrinivasanAnand opened 1 week ago

SrinivasanAnand commented 1 week ago

I'm working with the EndoDAC model to create depth estimation maps, and I am curious as to the units of the data in the _disp.npy data. I would like to confirm that the model output is on the same scale as the endoscopic video I am working with. The values in the npy array range from ~2 to ~9. Are these values in centimeters? If they are not currently representative of a standard unit, is there a way to convert them to a standard distance unit?

BeileiCui commented 5 days ago

Hi, first of all, disp refers to disparity so it is the reciprocal of depth. And our method is SSL method, so the results are relative disp (or depth). There are some ways to align the resulting depth to standard distance unit (or so called absolute depth). For example, before evaluating, we can either use a factor to multiply the estimated depth( which is obtained by the ratio of median of estimated depth and gt depth), or align both the scale and shift. You can check on paper and many other papers how to convert.