Awesome

PANet: Patch-Aware Network for Light Field Salient Object Detection

Introduction

Accepted paper in IEEE Trans on Cybernetics, 'PANet: Patch-Aware Network for Light Field Salient Object Detection', Yongri Piao, Yongyao Jiang, Miao Zhang, Jian Wang and Huchuan Lu.

Requirements

Windows 10

PyTorch 1.4.0

CUDA 10.0

Cudnn 7.6.0

Python 3.6.5

Numpy 1.16.4

Training

Modify your path of training dataset in config.py

Run train.py for training the saliency model, the maximum of training iterations is 500000.

Run train_mslm.py for training the MSLM model, the maximum of training iterations is 5000.

Run train_srm.py for training the SRM model, the maximum of training iterations is 5000.

Run train_second_decoder.py for training the second decoder, the maximum of training iterations is 500000.

Testing

Download pretrained models from here. Code: qwer

Modify your path of testing dataset in config.py

Run test to inference saliency maps

Saliency Maps

DUTS-LFSD&HFUT-LFSD&LFSD, Download link. Code: qwer

Contact and Questions

Contact:Yongyao Jiang. Email:572612808@qq.com