SALICON pytorch

This repository contains the code to train and run SALICON with pytorch. In our implementation we follow the original CVPR'15 paper

• Introduction
• Prerequisites
• Instruciton

Introduction

The implementation details of SALICON is based on Kotseruba's work, whichs is implemented with tensorflow

More detalis please see the original paper

Prerequisites

• Python>=3.5.0
• Pytorch>=1.5.0
• torchvision>=0.6.0
• opencv3>=3.1.0
• numpy>=1.14.2

Instruciton

Download Dataset

Download OSIE dataset if you want to train SALICON.

cd osie_dataset
sh download_osie_dataset.sh

Download MIT1003 dataset used for evaluation (optional).

cd mit1003_dataset
sh download_mit1003.sh

Train

1 . Train a model with default parameters.

python3 train_net.py

2 . You can change the tranning parameters according to your needs, such as whether to use GPU, etc.

# use gpu
python3 train_net.py --gpu --batch_size 1 --lr 0.1 --momentum 0.9 --decay 0.0005 --epochs 500 --checkpoint 1
# use cpu only
python3 train_net.py --batch_size 1 --lr 0.1 --momentum 0.9 --decay 0.0005 --epochs 500 --checkpoint 1

Test

1 . Test the model with default dataset (osie) .

python3 test.py

2 . Test the model with optional parameters, such as selecting the mit1003 dataset, whether to use gpu, etc.

python3 test.py --test_dataset mit1003 --gpu

Download the Model weights

The model weights must be at the same level as test.py in the src directory

1. Baidu NetDisk

https://pan.baidu.com/s/1IdWwChDfLmOpRZfyABbdIQ 
password: 6wpu

2. ecloud NetDisk

https://cloud.189.cn/t/ZZrumuBFnyia
password: 5jzf

3. Google Drive

https://drive.google.com/file/d/1hxa1rlm94cV_dABu-1B3Jg3cQ9j_f3-P/view?usp=sharing

Author

• Zhengxi Hu

Please raise an issue or send email to [email protected] if there are any issues running the code.