Skip to content

Latest commit

 

History

History
29 lines (24 loc) · 1.78 KB

DATASETS.md

File metadata and controls

29 lines (24 loc) · 1.78 KB

We use Gibson and Matterport3D datasets for our experiments. Please find below the instructions to download them.

Gibson

URL: http://gibsonenv.stanford.edu/database/
Target dir: $PONI_ROOT/data/scene_datasets/gibson_semantic

  • Download the Gibson scenes by agreeing to the terms of use.
  • Semantic annotations for Gibson is available from the 3DSceneGraph dataset.
  • Convert the semantic annotations to habitat format following instructions in habitat-sim.

Matterport3D dataset

URL: https://niessner.github.io/Matterport/
Target dir: $PONI_ROOT/data/scene_datasets/mp3d

  • Download the MP3D scenes by agreeing to the terms of use and using the offical Matterport3D download script. Note that this gives basis-compressed versions of the MP3D scenes, which supports fast simulation with efficient memory use.
    python download_mp.py --task_data habitat -o $PONI_ROOT/data/scene_datasets/mp3d
    
  • For extracting semantic maps, we require the non-basis compressed versions of the MP3D scenes. This can be obtained by downloading the raw MP3D scenes and converting them to GLB.
    python download_mp.py --type matterport_mesh -o $PONI_ROOT/data/scene_datasets/mp3d_uncompressed
    
  • Conversion to GLB can be performed using trimesh. Note: The *.ply files from $PONI_ROOT/data/scene_datasets/mp3d can be copied over to $PONI_ROOT/data/scene_datasets/mp3d_uncompressed.
    import trimesh
    scene = trimesh.load(obj_file_path)
    scene.export(glb_file_path)