- ABSNet: Aesthetics-Based Saliency Network using Multi-Task Convolutional Network(IEEE Signal Processing Letters2020)
- @ARTICLE{ABSNet,
author={Jing Liu, Jincheng Lv, Min Yuan, Jing Zhang, and Yuting Su},
journal={IEEE Signal Processing Letters},
title={Aesthetics-Based Saliency Network using Multi-Task Convolutional Network},
year={2020},
volume={},
number={},
pages={},
doi={10.1109/LSP.2020.3035065} }
- opencv-python 4.1.1
- numpy 1.16.1
- tenserflow 1.14.0
- tensorboard 1.14.0
- keras 2.2.2
- scipy 1.2.1
Download the following datasets and unzip them into data
folder.
Download pre-trained model BaiduYun (pwd: 4jk1).
This pre-trained model is for multi-task dalited resnet(MT-DRN), you should set 'weightfile' correctly.
- For testing sal branch, you should set config.yaml correctly and load pretrained pkl correctly, and then run main_sal_drn.py.
https://github.com/cvzoya/saliency
- SALICON Validation Set: Link:https://pan.baidu.com/s/1dPkwMpotQQhSQql4r42TFQ pwd: oxgw
- SALICON Test Set: Link: https://pan.baidu.com/s/1FygM6C1Su_6bcCivBW3QQw pwd: 6yqq
- MIT1003 DataSet: Link: https://pan.baidu.com/s/1z-q9BLrRBql2AZUUS_5mnA pwd: cwgx