# EBSR: Enhanced Binary Neural Network for Image Super-Resolution

**Authors:** Renjie Wei, Shuwen Zhang, Zechun Liu, Meng Li, Yuchen Fan, Runsheng Wang, Ru Huang  
**arXiv:** [2303.12270](https://arxiv.org/abs/2303.12270)  
**Date:** 2023-03-22  
**Summary page:** https://www.summarizepaper.com/en/paper/2303.12270v1/

## Key points

- Paper title: "EBSR: Enhanced Binary Neural Network for Image Super-Resolution"
- Authors: Renjie Wei, Shuwen Zhang, Zechun Liu, Meng Li, Yuchen Fan, Runsheng Wang, Ru Huang
- Challenge addressed: Deploying deep convolutional neural networks for image super-resolution on resource-constrained devices
- Spatial re-scaling
- Channel-wise shifting and re-scaling to augment binary convolutions and retain important spatial and channel-wise information
- Superior performance of EBSR models compared to prior art methods in terms of Peak Signal-to-Noise Ratio (PSNR) across different datasets and model sizes
- Promising solution for enhancing binary neural networks in image super-resolution tasks while improving model inference efficiency on resource-constrained devices

## Summary

The paper "EBSR: Enhanced Binary Neural Network for Image Super-Resolution" by Renjie Wei, Shuwen Zhang, Zechun Liu, Meng Li, Yuchen Fan, Runsheng Wang, and Ru Huang presents a solution to the challenge of deploying deep convolutional neural networks for image super-resolution (SR) on resource-constrained devices. The authors propose two methods - spatial re-scaling and channel-wise shifting and re-scaling - to augment binary convolutions and retain important spatial and channel-wise information crucial for achieving high-quality results. The proposed EBSR models demonstrate superior performance compared to prior art methods in terms of Peak Signal-to-Noise Ratio (PSNR) both quantitatively and qualitatively across different datasets and model sizes. This approach offers a promising solution for enhancing binary neural networks in image super-resolution tasks while improving model inference efficiency on resource-constrained devices.