HPSv2.1 is a metric for evaluating the human preference score of generated images. It learns human preference through fine-tuning CLIP model with 798K human ranked image pairs. The human ranked image pairs are from human experts.
| Methods | # Params | Painting | Photo | Animation | Concept-art | Average |
|---|---|---|---|---|---|---|
| Infinity | 2.20B | 30.45 | 29.33 | 31.66 | 30.41 | 30.46 |
| + ScaleKV | 2.20B | 30.32 | 29.26 | 31.54 | 30.21 | 30.33 |
| FastVAR | 2.20B | 28.47 | 27.61 | 29.77 | 25.58 | 28.61 |
pip install hpsv2
pip install -U diffusers
sudo apt install python3-tk
wget https://dl.fbaipublicfiles.com/mmf/clip/bpe_simple_vocab_16e6.txt.gz
# mv bpe_simple_vocab_16e6.txt.gz /home/tiger/.local/lib/python3.9/site-packages/hpsv2/src/open_clip
# Remember change your virtual environment name in `{your_vir-env_name}` and python version in `{python3.11}`
mv bpe_simple_vocab_16e6.txt.gz ~/anaconda3/envs/{your_vir-env_name}/lib/{python3.11}/site-packages/hpsv2/src/open_clip由于需要从Huggingface中下载文件，因此需先设置好 tools/conf.py 中的huggingface token。
运行脚本
bash scripts/hpsv2_infer4eval.sh- 测试结果可能与论文中不同，其他人也存在类似的问题，详情请见 FoundationVision/Infinity#42