---
license: apache-2.0
library_name: pytorch
tags:
- 3d
- visual-question-answering
- spatial-reasoning
- ground3d-lmm
---

# Ground3D-LMM — CA-VQA (3D+2D)

Fine-tuned **Ground3D-LMM** checkpoint for **CA-VQA** (paper **Table 9** — generality beyond ScanNet),
**3D+2D (point cloud + RGB views)** setting. Reproduces the paper's Table-9 average (~48).

- **Checkpoint file:** `pytorch_model.pth`
- **Code:** https://github.com/AmolHarsh/ground3d-lmm

## Usage
```bash
CKPT=$(huggingface-cli download amolharsh/Ground3D-LMM-CAVQA-4B-Joint pytorch_model.pth)
# evaluate with the CA-VQA harness: infer_cavqa.py -> merge_preds.py -> evaluate_cavqa.py
```