Articulate3D Challenge

Dataset Structure

Each scene contains two JSON files:

{scannetpp_scan_id}_parts.json
      {scannetpp_scan_id}_artic.json

• {scannetpp_scan_id}: ScanNet++ scene ID
• parts.json: Part segmentation annotations
• artic.json: Articulation (motion) annotations

Part Segmentation

• Face-based segmentation via triIndices
• Vertex labels can be derived by voting over faces
• Hierarchy encoding:

  {obj_id}.{parent_hierarchy}.{own_hierarchy}.{label}

Example:

3.1.cabinet
      3.1.2_1.door
      3.1.2_2.door
      3.1.2_1.3_1.handle

• 3.1.cabinet: Cabinet object
• 3.1.2_1.door, 3.1.2_2.door: Cabinet doors
• 3.1.2_1.3_1.handle: Handle of the first door

Articulation Annotations

• Each movable part is identified by a pid
• pid corresponds to partId in parts.json
• The base field indicates the static reference part
• The base can also be inferred from the hierarchy

Movable Parts & Articulation (.zip)

{
      "scene_id_1": {
        "pred_masks": numpy.Array,
        "pred_scores": numpy.Array,
        "pred_classes": numpy.Array,  // 1: rotation, 2: translation
        "pred_origins": numpy.Array,
        "pred_axes": numpy.Array
      },
      ...
    }

Interactable Parts (.pkl)

{
      "scene_id_1": {
        "pred_masks": numpy.Array,
        "pred_scores": numpy.Array,
        "pred_classes": numpy.Array  // articulation type of parent movable part
      },
      ...
    }