search

apple / ml-depth-pro

Depth Pro: Sharp Monocular Metric Depth in Less Than a Second.
Other
3.73k stars 250 forks source link

Can we use this for classification? #32

Open nhha1602 opened 1 month ago

nhha1602 commented 1 month ago

Hi,

Can we use this for classification: real image that captured from camera and spoof image that captured from image displayed on screen of PC/PAT/Phone ?

Thank you very much.

99991 commented 1 month ago

Maybe. I computed the depth map for a photo of a phone showing an image of cars and the depth map is flat where the phone is:

photo

So you could run some image detection algorithm to find the location of the subject/object you are looking for and then check whether the average norm of gradient of the depth map is small in that region. If the gradient is small, then the object is probably shown on a screen. But this solution might not be robust, for example a better screen might fool the network.