Objaverse -- LVIS subset draw

objaverse_lvis · Objaverse 1.0 · planned — not present locally yet

Ground-truth mesh

Not present locally. See the acquisition plan below.

Why this sample is in the benchmark

Breadth over depth: sample many categories to check that a conclusion holds beyond hand-picked objects. Mesh quality varies, which is itself a useful stress axis (non-manifold and open meshes are common).

Ground truth

GT type

authored

licence

Per-object (CC0 / CC-BY / CC-BY-NC) -- check metadata

source

Objaverse 1.0

builder

- {'uid': 'TODO_pick_object'}

views

40 × 512px (orbit, fov 50°)

lighting

provided

Acquisition plan

kind api, automatic False, estimated size 25 MiB

1. Install the downloader

   pip install objaverse

2. Fetch the object and its annotations

   python -c "import objaverse; o=objaverse.load_objects(['TODO_pick_object']); print(o)"

3. Explore the LVIS subset (category-balanced)

   python -c "import objaverse; a=objaverse.load_annotations(); print(len(a))"

4. Object page / dataset home
   https://objaverse.allenai.org/

Objaverse 1.0: 800k+ annotated objects. Per-object licences vary (CC-BY, CC-BY-NC, CC0); always read the object's metadata before redistributing. Use the LVIS subset for a balanced, category-diverse draw.

Comparisons

No reconstruction compared yet. Run geombench compare objaverse_lvis --mock or point it at a real 3DGS output.