facebookresearch / facebookresearch/sam-3d-objects

Evaluation Data: IM-MA mesh geometry defect breakdown across 101 exports

Open
#201 0 comments 0 reactions 0 assignees View on GitHub
Dominant language
Python
Stars
7.4k
Forks
878
PR merge metrics
No merged PRs in 30d

Description

Hi IM-3D / Meta GenAI team,

I ran an open, reproducible geometry QA census (`3dqa-benchmark`) across 2,307 generative exports from 23 models.

**IM-MA Metrics (n=101):**
- Non-watertight: 84.2%
- Non-manifold edges: 53.5%
- Missing UVs: 100%
- Vertex-colour-only exports: 100%
- Exceeding 30k budget: 0.0%

Full reproducible CLI runner (`pip install 3dqa`) and aggregate reports are available at:
https://github.com/alza123123/3dqa-benchmark

Contributor guide

Open the contributing guide

Research direction

Start with the linked 3dqa-benchmark CLI runner and its aggregate reports to reproduce the IM-MA measurements. The issue names no repository files, tests, or specific requested change; completion would require clarifying which geometry defect to address and how the 101 exports should be evaluated.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
computer-graphics, testing-qa
Issue type
Bug
Difficulty
5/5
Estimated time
Over a week
Activity status
Quiet
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.