SalNAS: Efficient Saliency-prediction Neural Architecture Search with self-knowledge distillation

This paper has been published to Engineering Applications of Artificial Intelligence.

Paper: EAAI version or arXiv version

Overview

This repository contains the source code for SalNAS, which accompanies the research paper titled SalNAS: Saliency-Prediction Neural Architecture Search with Self-Knowledge Distillation. The purpose of this repository is to provide transparency and reproducibility of the research results presented in the paper.

This code is based on the implementation of EML-NET-Saliency, SimpleNet, MSI-Net, EEEA-Net, and AlphaNet.

Prerequisite for server

• Tested on Ubuntu OS version 22.04 LTS
• Tested on Python 3.11.8
• Tested on CUDA 12.3
• Tested on PyTorch 2.2.1 and TorchVision 0.17.1
• Tested on NVIDIA RTX 4090 24 GB

Cloning source code

git clone https://github.com/chakkritte/SalNAS/
cd SalNAS
mkdir data

The dataset folder structure:

PKD
|__ data
    |_ salicon
      |_ fixations
      |_ saliency
      |_ stimuli
    |_ mit1003
      |_ fixations
      |_ saliency
      |_ stimuli
    |_ cat2000
      |_ fixations
      |_ saliency
      |_ stimuli
    |_ pascals
      |_ fixations
      |_ saliency
      |_ stimuli
    |_ osie
      |_ fixations
      |_ saliency
      |_ stimuli

Creating new environments

conda create -n salnas python=3.11.8
conda activate salnas
conda install pytorch torchvision torchaudio pytorch-cuda=12.1 -c pytorch -c nvidia

Install Requirements

pip install -r requirements.txt --no-cache-dir

Usage

Training SalNAS supernet on Salicon dataset

python train_salnas.py --amp --self_kd --no_epochs 20 --t_epochs 10  --loss_mode new --kldiv --cc --nss --output_dir output-selfkd

Citation

If you use SalNAS or any part of this research, please cite our paper:

@article{termritthikun2024salnas,
  title = "{SalNAS: Efficient Saliency-prediction Neural Architecture Search with self-knowledge distillation}",
  journal = {Engineering Applications of Artificial Intelligence},
  volume = {136},
  pages = {109030},
  year = {2024},
  issn = {0952-1976},
  doi = {https://doi.org/10.1016/j.engappai.2024.109030},
  author = {Chakkrit Termritthikun and Ayaz Umer and Suwichaya Suwanwimolkul and Feng Xia and Ivan Lee},
}

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.