search

alexklwong / mondi-python

PyTorch Implementation of Monitored Distillation for Positive Congruent Depth Completion (ECCV 2022)
43 stars 3 forks source link

Running in real-time on live data #7

Open Srijal97 opened 9 months ago

Srijal97 commented 9 months ago

Hello! I'm able to set up and run MonDi on the VOID dataset and get evaluation results similar to what is presented.

Evaluation results:
     MAE      RMSE      iMAE     iRMSE
  30.884    87.478    15.308    38.333
     +/-       +/-       +/-       +/-
  23.885    79.675    18.747    51.171
Total time: 13383.36 ms  Average time per sample: 16.73 ms

How could I go about running MonDi in real-time on live images? I noticed that the dataset images are preprocessed to be in triplets. Would this have to be done for real-time images too? Also, my depth stream is as a pointcloud; would that have to be saved as PNG depth images as in the dataset?

Thank you!

edhyah commented 7 months ago

@alexklwong Hey Alex, similar question - is it possible to feed it dense depth maps? I'm feeding it a dense depth map from an Azure Kinect and am struggling to see good results, but not sure if it's because it's dense or there's some other parameter I'm inputting wrong.

alexklwong commented 7 months ago

Hi, it looks like the code base is an old version of the original (in fact, I looked at the training loop and it looks like there is an error). I will ask the students to prepare the new code and add an online inference script.