/Gibson_Spares_Depth_Completion

Gibson_Spares_Depth_Completion

Primary LanguagePython

Gibson+Sparse-Depth-Completion

This repo first create a virtual robotics environment using Gibson. RGB and spare LiDAR data are generated in the Gibson. Then I use the trained model of Sparse-Depth-Completion (https://github.com/wvangansbeke/Sparse-Depth-Completion) to perform guided up sampling and generate a dense depth image. The training runs on HIPERGATOR 3.0.

demo demo

Run the project

  1. Install Gibson. There are several options. You can download the Gibson, or follow http://svl.stanford.edu/igibson/docs/installation.html.

  2. Run iGibson/demo5.py to generate RGB images and LiDAR depth images.

  3. Train the model, python dkmain.py. Refer to the [Sparse-Depth-Completion] (https://github.com/wvangansbeke/Sparse-Depth-Completion) for the training data.

  4. Run Test/test.py to generate the dense depth image.

  5. Use creatVideo.py to generate videos.