MESFINet

Official Pytorch implementation of "Multi-Stage Edge-Guided Stereo Feature Interaction Network for Stereoscopic Image Super-Resolution" paper.

Overview

The architecture of our proposed Multi-Stage Edge-Guided Stereo Feature Interaction Network.

Citation:

@ARTICLE{10121360,
  author={Wan, Jin and Yin, Hui and Liu, Zhihao and Liu, Yanting and Wang, Song},
  journal={IEEE Transactions on Broadcasting}, 
  title={Multi-Stage Edge-Guided Stereo Feature Interaction Network for Stereoscopic Image Super-Resolution}, 
  year={2023},
  volume={69},
  number={2},
  pages={357-368},
  doi={10.1109/TBC.2023.3264880}}

Train
Test
Results

Train

Begin to train

Run the following scripts to train models. You can use scripts in the file 'demo' to train models for our paper.

# BI, scale 2, 4
# MESFINet in the paper (x2)
# CUDA_VISIBLE_DEVICES=3,4 python train.py --model_name MESFINet --scale_factor 2 --checkpoints_dir ./log \
# --edge_model_path ./pretrained_model/bdcn_pretrained_on_nyudv2_rgb.pth 

# MESFINet in the paper (x4) - from MESFINet (x2)
# CUDA_VISIBLE_DEVICES=3,4 python train.py --model_name MESFINet --scale_factor 4 --checkpoints_dir ./log --edge_model_path ./pretrained_model/bdcn_pretrained_on_nyudv2_rgb.pth \
# --load_pretrain_modelx2 True --modelx2_path ./pretrained_model/MESFINet_2xSR_final.pth.tar

Test

Clone this repository:

git clone https://github.com/jinwan1994/MESFINet.git

All the models (BIX2/4) can be downloaded from GoogleYun, place the models to ./pretrained_model/.

Run the following scripts.

You can use scripts in the file 'demo' to produce results for our paper.

# BI, scale 2, 4
# Standard benchmarks (Ex. MESFINetx2)
# python test.py --model_name MESFINet_2xSR --scale_factor 2 --save_dir ./results_test --sr_model_path ./pretrained_model/MESFINet_2xSR_final.pth.tar \
# 	--dataset_list KITTI2012+KITTI2015+Middlebury --edge_model_path ./pretrained_model/bdcn_pretrained_on_nyudv2_rgb.pth

# Standard benchmarks (Ex. MESFINet_x4)
python test.py --model_name MESFINet_4xSR --scale_factor 4 --save_dir ./results_test --sr_model_path ./pretrained_model/MESFINet_4xSR_final.pth.tar \
	--dataset_list KITTI2012+KITTI2015+Middlebury --edge_model_path ./pretrained_model/bdcn_pretrained_on_nyudv2_rgb.pth

Finally, SR results and PSNR/SSIM values for test data are saved to ./results_test/*. (PSNR/SSIM values in our paper are obtained using Matlab2021)

Results

Quantitative Results

Benchmark SR results. Average PSNR/SSIM for scale factor x2 and x4 on four datasets.

Visual Results

Visual comparison for 2x SR on two datasets.

jinwan1994/MESFINet