Portrait-Quality-Assessment

This repository contains the code to train and evaluate different models and methods for portrait quality assessment for PIQ23 Dataset (CVPR 2023).

Dataset

• For original dataset, please refer to PIQ23 Dataset (CVPR 2023).
• After downloading the dataset, please put images in PIQ23/Images folder.

Training

This code is config-based. You can modify the config file to train different models and methods. Example:

python train.py --config configs/resnet/resnet18.yaml

Specific config will be automatically merged with configs/base.yaml.

Results

Now supports following backbone:

Model	Freeze	Val acc
Resnet18	False
Resnet50	False

TODO

• Add more backbones
• Add more methods

Reference

Thanks to authors of PIQ23 Dataset (CVPR 2023), they have done a great job!