face image quality assessment ediffiqa
简介
核心亮点
- 快速量化人脸图像质量，有效过滤模糊低质照片
- 适配人脸识别预处理，提升后端算法识别率
- 基于 OpenCV 生态，集成简单且部署成本极低
- 采用 Apache-2.0 协议，支持商业化灵活部署
使用方法
# 安装 Hugging Face transformers
pip install transformers torch# 使用 transformers 加载模型
from transformers import AutoModel, AutoTokenizer
model = AutoModel.from_pretrained("opencv/face_image_quality_assessment_ediffiqa")
tokenizer = AutoTokenizer.from_pretrained("opencv/face_image_quality_assessment_ediffiqa")Hugging Face 下载
我们推荐使用命令行或者 Hugging Face Hub SDK 来进行模型的下载。
操作指引：在下载前，请先通过如下命令安装 huggingface_hub：
pip install -U huggingface_hub命令行下载
下载完整模型库
huggingface-cli download opencv/face_image_quality_assessment_ediffiqa下载单个文件到指定本地文件夹（以下载 config.json 到当前路径下 ./dir 目录为例）
huggingface-cli download opencv/face_image_quality_assessment_ediffiqa config.json --local-dir ./dirSDK 下载
# 模型下载
from huggingface_hub import snapshot_download
model_dir = snapshot_download('opencv/face_image_quality_assessment_ediffiqa')Git 下载
请确保 lfs 已经被正确安装
git lfs install
git clone https://huggingface.co/opencv/face_image_quality_assessment_ediffiqa如果您希望跳过 lfs 大文件下载，可以使用如下命令
GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/opencv/face_image_quality_assessment_ediffiqa模型文件托管在 Hugging Face Hub，使用 HF CLI / SDK / Git 直接下载，不经过本站。
PyTorch / Transformers 使用
安装 Transformers
pip install -U transformers torch模型加载和推理
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained('opencv/face_image_quality_assessment_ediffiqa')
tokenizer = AutoTokenizer.from_pretrained('opencv/face_image_quality_assessment_ediffiqa')完整文档
eDifFIQA(T)
eDifFIQA(T) is a light-weight version of the models presented in the paper eDifFIQA: Towards Efficient Face Image Quality Assessment based on Denoising Diffusion Probabilistic Models, it achieves state-of-the-art results in the field of face image quality assessment.
Notes:
- The original implementation can be found here.
- The included model combines a pretrained MobileFaceNet backbone, with a quality regression head trained using the proceedure presented in the original paper.
- The model predicts quality scores of aligned face samples, where a higher predicted score corresponds to a higher quality of the input sample.
- In the figure below we show the quality distribution on two distinct datasets: LFW[[1]](#1) and XQLFW[[2]](#2). The LFW dataset contains images of relatively high quality, whereas the XQLFW dataset contains images of variable quality. There is a clear difference between the two distributions, with high quality images from the LFW dataset receiving quality scores higher than 0.5, while the mixed images from XQLFW receive much lower quality scores on average.
<a id="1">[1]</a>
B. Huang, M. Ramesh, T. Berg, and E. Learned-Miller
“Labeled Faces in the Wild: A Database for Studying Face Recognition in Unconstrained Environments”
University of Massachusetts, Amherst, Tech. Rep. 07-49,
October 2007.
<a id="2">[2]</a>
M. Knoche, S. Hormann, and G. Rigoll
“Cross-Quality LFW: A Database for Analyzing Cross-Resolution Image Face Recognition in Unconstrained Environments,” in Proceedings of the IEEE International Conference on Automatic Face and Gesture Recognition (FG), 2021, pp. 1–5.
Demo
*NOTE*: The provided demo uses ../face_detection_yunet for face detection, in order to properly align the face samples, while the original implementation uses a RetinaFace(ResNet50) model, which might cause some differences between the results of the two implementations.
To try the demo run the following commands:
# Assess the quality of 'image1'
python demo.py -i /path/to/image1
Output all the arguments of the demo
python demo.py --helpExample outputs
The demo outputs the quality of the sample via terminal (print) and via image in __results.jpg__.
License
All files in this directory are licensed under CC-BY-4.0.
