DeQA-Score-Mix3

DeQA-Score ( project page / codes / paper ) model weights fully fine-tuned on KonIQ, SPAQ, and KADID datasets.

This work is under our DepictQA project.

Quick Start with AutoModel

For this image, start an AutoModel scorer with transformers==4.36.1:

import requests
import torch
from transformers import AutoModelForCausalLM

model = AutoModelForCausalLM.from_pretrained(
  "zhiyuanyou/DeQA-Score-Mix3", trust_remote_code=True, torch_dtype=torch.float16, device_map="auto"
)

from PIL import Image

# The inputs should be a list of multiple PIL images
score = model.score(
  [Image.open(requests.get(
    "https://raw.githubusercontent.com/zhiyuanyou/DeQA-Score/main/fig/singapore_flyer.jpg", stream=True
    ).raw)]
)

The "score" result should be 1.9404 (in range [1,5], higher is better).

Non-reference IQA Results (PLCC / SRCC)

Dataset	KonIQ	SPAQ	KADID	PIPAL	LIVE-Wild	AGIQA	TID2013	CSIQ
Q-Align (Baseline)	0.945 / 0.938	0.933 / 0.931	0.935 / 0.934	0.409 / 0.420	0.887 / 0.883	0.788 / 0.733	0.829 / 0.808	0.876 / 0.845
DeQA-Score (Ours)	0.956 / 0.943	0.938 / 0.934	0.955 / 0.953	0.495 / 0.496	0.900 / 0.887	0.808 / 0.745	0.852 / 0.820	0.900 / 0.857

If you find our work useful for your research and applications, please cite using the BibTeX:

@article{deqa_score,
  title={Teaching Large Language Models to Regress Accurate Image Quality Scores using Score Distribution},
  author={You, Zhiyuan and Cai, Xin and Gu, Jinjin and Xue, Tianfan and Dong, Chao},
  journal={arXiv preprint arXiv:2501.11561},
  year={2025},
}