mifanfan/MesaTask-CTRC-100-target

Tabletop Scene Restoration Dataset - Target Layouts

This dataset holds the target state of a scene restoration task: each scene should be restored to the arrangement given here.

source.scene on each object identifies its scene and can be used to pair with other data from the same batch.

• Scenes: 100
• Scene types: bathroom_vanity 22, coffee_table 9, dining_table 11, dressing_table 24, kitchen_counter 10, office_table 24
• Objects moved per scene: 1-3

Directory layout

scenes/<scene_type>/<level>/<scene_name>/
├── layout.json        scene layout of this dataset
├── render.png         render of this layout
└── meta.json          list of moved objects and validation record
assets/              all GLB models referenced by this dataset
manifest.json        dataset listing and statistics

The dataset is self-contained: every referenced model lives under assets/, there are no absolute paths, and no external asset library is required.

Layout JSON format

{
  "scene_settings": {
    "units": "centimeters",
    "up_axis": "Z",
    "rotation_convention": "absolute_quaternion_xyzw",
    "scene_offset": [x, y, z]   // meters, see below
  },
  "item_placement_zone": [xmin, xmax, ymin, ymax],   // centimeters, table extent
  "objects": [
    {
      "instance": "3_bowl_0",
      "position": [x, y, z],          // cm, model origin, not bbox center
      "size": [sx, sy, sz],           // cm, (x, y) horizontal extents, z height
      "rotation": [x, y, z, w],       // absolute quaternion
      "resolved_glb": "<uid>.glb",    // model file to load from assets/
      "selected_uid": "<uid>"
    }
  ]
}

Notes:

• position is the model origin (not the bounding-box center), in centimeters
• size is the object's target extent; the model is already scaled to match it. (x, y) are the horizontal extents and z is the height. A model's own local up-axis may not align with z, so orient it using size as the reference
• rotation is the complete absolute orientation as a quaternion in [x, y, z, w] order. Apply it directly, per object, and do not compose any additional rotation on top
• When an object rests flat on the table, position[2] is approximately size[2] / 2

What is scene_offset?

Each object's position is given in the scene's own local frame, whose origin is one corner of the table (the lower-left corner of item_placement_zone), in centimeters. Using that frame as-is would stack every scene's table in the same place.

scene_settings.scene_offset (in meters) is the translation that moves these local coordinates into a shared world frame:

world_x = position_cm[0] * 0.01 + scene_offset[0]
world_y = position_cm[1] * 0.01 + scene_offset[1]
world_z = position_cm[2] * 0.01 + scene_offset[2]

In this dataset the effect is: the table is centered on the origin along x, and its near long edge sits at y = 0.10 m, i.e. the table occupies x in [-W/2, +W/2] and y in [0.10, 0.10 + D], where W and D are the table's horizontal and depth span in item_placement_zone. The z component is always 0.

Notes:

• It is only a suggested placement. You may ignore it and lay each scene out in its own local frame instead; the relative arrangement of objects inside a scene is unaffected
• scene_offset is a pure translation - no rotation, no scaling
• For an object resting flat on the table, world_z is its height above the table, and the tabletop corresponds to world_z = 0

Renders

Renders use the native layout coordinate frame. They are meant only for visually checking scene contents and object placement.