y-zheng18 / point_odyssey

Official code for PointOdyssey: A Large-Scale Synthetic Dataset for Long-Term Point Tracking (ICCV 2023)
Other
131 stars 5 forks source link

Questions regarding depth values #14

Closed lukaboljevic closed 4 months ago

lukaboljevic commented 4 months ago

Hello! I have two questions regarding the depth images/values.

Based to my understanding, the ground truth depth PNG images are 16-bit. Then, by dividing with 65535.0 and multiplying with 1000.0 (as in this line) the depth is converted to milimeters? Is this correct? I inferred this from the other two issues related to depth (#2, #6), but these lines then confuse me:

https://github.com/y-zheng18/point_odyssey/blob/2e8cc2f75b78b966da279bc76ba171ff42b732ca/utils/reprojection.py#L47-L49

It states that you want to use max depth of 10 m(eters) for visualization. Can you clarify?

Secondly, what is the "unit" for the original depth values (from the 16-bit PNG image), i.e. what do the values represent? Obviously, depth, but in meters? Milimeters? Or something else?

Sorry if the questions are silly, it's my first time dealing with depth images so I wanted to understand what I'm looking at. Thanks in advance!

y-zheng18 commented 4 months ago

Hi, the original depth is rendered as .tiff format in meters, and converted to png by using (depth_tiff 65535 / 1000), supposing the max depth is 1000 meters. To convert png back to meter depth, you can use depth_png / 65535 1000.

lukaboljevic commented 4 months ago

Ah, so everything is in meters, I see. It makes more sense now. That answers both questions then. Thank you very much for the quick answer!