Home

Awesome

MPADA

The implementation of $MP_{ada}$ in Attention-based Multi-patch Aggregation for Image Aesthetic Assessment pdf, the method for SOTA aesthetic visual assessment performance on AVA benchmark. For more comparisons on AVA, please refer to the page on PaperWithCode.

Framework

<p align="center"> <img src="https://github.com/Openning07/MPADA/blob/master/FromPaper/SystemOverview.png" alt="CMM" width="52%"> </p>

System overview. We use an attention-based objective to enhance training signals by assigning relatively larger weights to misclassified image patches.

Experiments

Requirements

Notes

Instructions for Results in the paper

python AVA2012-resnet_20180808_Revised.py --gpu 2 --data $YOUR_DATA_DIR$/AVA2012
        --aesthetic_level 2 --crop_method_TS RandomCrop --repeat_times 15
        --load $YOUR_CHECKPOINT_DIR$/checkpoint --mode resnet -d 18 --eval 

Desired Outputs

TODO

Notes

Citation

Please cite the following paper if you use this repository in your reseach~ Thank you ^ . ^

@inproceedings{sheng2018attention,
  title={Attention-based multi-patch aggregation for image aesthetic assessment},
  author={Sheng, Kekai and Dong, Weiming and Ma, Chongyang and Mei, Xing and Huang, Feiyue and Hu, Bao-Gang},
  booktitle={2018 ACM Multimedia Conference on Multimedia Conference},
  pages={879--886},
  year={2018},
  organization={ACM}
}