It was fortunate to find the missing annotated data for the GAIC dataset on my old laptop (see directory "GAIC-Text-Annotations").
Hopefully these annotations will be useful to the field of smart cropping.
I personally don't follow this field, so there will be no updates.
Download the GAIC dataset (journal version) from here, which includes 2636, 200, and 500 images for training, validation and testing.
If you find this repository useful, please consider citing:
@inproceedings{zhong2023clipcrop,
title={ClipCrop: conditioned cropping driven by vision-language model},
author={Zhong, Zhihang and Cheng, Mingxi and Wu, Zhirong and Yuan, Yuhui and Zheng, Yinqiang and Li, Ji and Hu, Han and Lin, Stephen and Sato, Yoichi and Sato, Imari},
booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
pages={294--304},
year={2023}
}