search

isl-org / DPT

Dense Prediction Transformers
MIT License
1.96k stars 254 forks source link

Depth output unit #72

Open CristianViSa opened 2 years ago

CristianViSa commented 2 years ago

Hi i have a question. I wanted to know what is the units of the output depth map given.

For an image, using kitti pretrained,i obtain for a determined pixel a value of 11, but using NYU the value is 3 and using MIDAS the value is 30. I just wanted to know, what is this?

Is it meters (dont think so), units, etc. Thanks

lizelive commented 1 year ago

yeah im curious about this as well

v-parmar commented 1 year ago

Hey @CristianViSa, The unit of the output depth map depends on the specific depth prediction algorithm used. For the depth prediction models trained on the KITTI dataset, the output depth map typically represents the estimated inverse depth of each pixel. The same is likely true for the models trained on the NYU and MIDAS datasets, but it would depend on the specific training data and loss functions used. Without more information on the exact models you're using, it's difficult to determine the exact unit of the output depth maps.