# synbench

**Repository Path**: mirrors_ibm/synbench

## Basic Information

- **Project Name**: synbench
- **Description**: Official repo for the paper "SynBench: Task-Agnostic Benchmarking of Pretrained Representations using Synthetic Data"
- **Primary Language**: Unknown
- **License**: Apache-2.0
- **Default Branch**: main
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2022-10-24
- **Last Updated**: 2025-09-28

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# synbench
Official repo for the paper "SynBench: Task-Agnostic Benchmarking of Pretrained Representations using Synthetic Data"

### Enviroment setup
```
pip install -r requirements.txt
```

### Quickstart 
Evaluate 'vit_base_patch16_224' with SynBench
```
python evaluate_synbench.py --fm_name vit_base_patch16_224 --calculate 1
```
### Cite
```
@article{ko2022synbench,
  title={SynBench: Task-Agnostic Benchmarking of Pretrained Representations using Synthetic Data},
  author={Ko, Ching-Yun and Chen, Pin-Yu and Mohapatra, Jeet and Das, Payel and Daniel, Luca},
  journal={arXiv preprint arXiv:2210.02989},
  year={2022}
}
```