# xUnit: Learning a Spatial Activation Function for Efficient Image Restoration

arXiv: https://arxiv.org/abs/1711.06445 · HTML: https://mlsanity.com/paper/1711.06445/

## Key facts

- Cited by 10 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 17 Nov 2017.

## Built on by (most cited first, 10 of 10)

- [Learning Normal Dynamics in Videos with Meta Prototype Network](https://mlsanity.com/paper/2104.06689/) (2021)
- [A Comprehensive Review of Deep Learning-based Single Image Super-resolution](https://mlsanity.com/paper/2102.09351/) (2021)
- [Dynamic Pre-training: Towards Efficient and Scalable All-in-One Image Restoration](https://mlsanity.com/paper/2404.02154/) (2024)
- [Why KDAC? A general activation function for knowledge discovery](https://mlsanity.com/paper/2111.13858/) (2021)
- [UnfairGAN: An Enhanced Generative Adversarial Network for Raindrop Removal from A Single Image](https://mlsanity.com/paper/2110.05523/) (2021)
- [Sparsity Aware Normalization for GANs](https://mlsanity.com/paper/2103.02458/) (2021)
- [SP3: Spherical Priors for Plug-and-Play Restoration](https://mlsanity.com/paper/2606.16396/) (2026)
- [Normal Learning in Videos with Attention Prototype Network](https://mlsanity.com/paper/2108.11055/) (2021)
- [Mixture-of-Rookies: Saving DNN Computations by Predicting ReLU Outputs](https://mlsanity.com/paper/2202.04990/) (2022)
- [PDC-ViT : Source Camera Identification using Pixel Difference Convolution and Vision Transformer](https://mlsanity.com/paper/2501.16227/) (2025)

## Abstract (arXiv)

In recent years, deep neural networks (DNNs) achieved unprecedented performance in many low-level vision tasks. However, state-of-the-art results are typically achieved by very deep networks, which can reach tens of layers with tens of millions of parameters. To make DNNs implementable on platforms with limited resources, it is necessary to weaken the tradeoff between performance and efficiency. In this paper, we propose a new activation unit, which is particularly suitable for image restoration problems. In contrast to the widespread per-pixel activation units, like ReLUs and sigmoids, our unit implements a learnable nonlinear function with spatial connections. This enables the net to capture much more complex features, thus requiring a significantly smaller number of layers in order to reach the same performance. We illustrate the effectiveness of our units through experiments with state-of-the-art nets for denoising, de-raining, and super resolution, which are already considered to be very small. With our approach, we are able to further reduce these models by nearly 50% without incurring any degradation in performance.