Home

Awesome

Official repository for CVPR2023 paper: "Towards Artistic Image Aesthetics Assessment: a Large-scale Dataset and a New Method"

Dataset

git clone https://github.com/Dreemurr-T/BAID.git
cd BAID/
pip install pandas
pip install tqdm
python downloading_script/download.py

The images will be saved to images/ folder.

Since it might be slow when downloading the images, we provide alternatives to obtain the dataset:

Ground-truth labels of the dataset can be found in the dataset folder.

Code

Requirements

Other dependencies can be installed with:

pip install -r requirements.txt

Pretraining

python pretraining_utils/pretrain_mani.py
python pretraining.py

The whole pretraining process takes about 2 days on a single RTX3090. We provide our pretrained weights at Drive.

Training

For training on BAID, use:

python train.py

Checkpoints will be save to checkpoint/SAAN folder.

Testing

For testing on BAID, download the pretrained weights from Drive, place the checkpoint in checkpoint/BAID

Then use:

python test.py

License

The dataset is licensed under CC BY-NC-ND 4.0

Acknowledgement

The code borrowed from pytorch-AdaIN and Non-local_pytorch.

Citation

If you find our work useful, please cite our work as:

@InProceedings{Yi_2023_CVPR,
    author    = {Yi, Ran and Tian, Haoyuan and Gu, Zhihao and Lai, Yu-Kun and Rosin, Paul L.},
    title     = {Towards Artistic Image Aesthetics Assessment: A Large-Scale Dataset and a New Method},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2023},
    pages     = {22388-22397}
}