ssecv / PSR

The official implementation of ACM MM 2023 "Partitioned Saliency Ranking with Dense Pyramid Transformers"

Partitioned Saliency Ranking with Dense Pyramid Transformers. ACM MM, 2023. Arxiv

Requirements

• Python $\ge$ 3.8
• PyTorch $\ge$ 1.9.0 and torchvision that matches the installation.
• setuptools == 59.5.0
• numpy == 1.23.0

Environment

The code is tested on CUDA 11.1 and pytorch 1.9.0, change the versions below to your desired ones. First install AdelaiDet and Detectron2 following the offical guide: AdelaiDet

Then build PSR with:

cd PSR
python setup.py build develop

Dataset

Download the datasets from the following links from original authors

• ASSR: Github
• IRSR: Github

Usage

Download

Pre-trained model weights are come from AdelaiDet:

Model	Config	Download
R50	config	model
R101	config	model

These are the weights trained on ASSR dataset:

• R50: Baidu Disk Verification code: 6z4u; Google Disk
• R101: Baidu Disk Verification code: qy4k; Google Disk

Train

python train_psr.py --config-file configs/R50_3x.yaml

Inference

python train_psr.py --config-file configs/R50_3x.yaml \
    --eval-only MODEL.WEIGHTS {PATH_TO_PRE_TRAINED_WEIGHTS}

Please replace {PATH_TO_PRE_TRAINED_WEIGHTS} to the pre-trained weights

Citation

@inproceedings{mm2023psr,
  title={Partitioned Saliency Ranking with Dense Pyramid Transformers},
  author={Sun, Chengxiao and Xu, Yan and Jialun, Pei and Fang, Haopeng and Tang, He},
  booktitle={Proceedings of the 31st ACM International Conference on Multimedia (MM '23), October 29-November 3, 2023, Ottawa, ON, Canada,
  year={2023}
}