---
language:
- en
license: cc-by-nc-4.0
task_categories:
- video-classification
- visual-question-answering
- video-text-to-text
pretty_name: RoboJudge-Bench
tags:
- robojudge
- embodied-ai
- video-evaluation
- benchmark
---

# RoboJudge-Bench

**Paper:** [RoboJudge: Multimodal Language Models as Judges for Embodied Video Generation](https://huggingface.co/papers/2610.27782)  
**Code:** [GitHub - Sphere-AI-Lab/RoboJudge](https://github.com/Sphere-AI-Lab/RoboJudge)  
**Project page:** https://siyuanmacs.github.io/RoboJudge/

RoboJudge-Bench contains 800 human-annotated embodied videos from 8 source corpora and 10 video generators. Every record includes the task instruction, locally hosted video and initial-frame URLs, PA/IA ratings, and six diagnostic sub-scores.

## Files

- `data/test.jsonl`: benchmark metadata, instructions, links, and labels.
- `data/<source>/generated_data/.../*.mp4`: all 800 benchmark videos.
- `data/<source>/generated_data/.../prompt/`: initial frames and instructions.
- `metadata/summary.json`: release counts and source revision.

## Scores

- **Physical Adherence (PA):** agent consistency, scene consistency, and interaction realism.
- **Instruction Alignment (IA):** agent match, object correctness, and goal completion.

Main scores are integers from 1 to 5. Diagnostic sub-scores are integers from 0 to 2.

## Sample Usage

```bash
python -m pip install -r requirements.txt
huggingface-cli download HuggingFriends/RoboJudge-Bench \
  --repo-type dataset --local-dir data_root

CKPT=HuggingFriends/RoboJudge-9B \
DATA_ROOT="$PWD/data_root" \
LIMIT=4 \
bash code/run_inference.sh

python code/evaluate.py \
  outputs/predictions.jsonl \
  data/benchmark/robojudge_bench.jsonl
```