# The Level Weighted Structural Similarity Loss: A Step Away from the MSE

arXiv: https://arxiv.org/abs/1904.13362 · HTML: https://mlsanity.com/paper/1904.13362/

## Key facts

- Cited by 6 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 30 Apr 2019.

## Built on by (most cited first, 6 of 6)

- [Robust Representation Learning via Perceptual Similarity Metrics](https://mlsanity.com/paper/2106.06620/) (2021)
- [Video Quality Assessment: A Comprehensive Survey](https://mlsanity.com/paper/2412.04508/) (2024)
- [Texture-enhanced Light Field Super-resolution with Spatio-Angular Decomposition Kernels](https://mlsanity.com/paper/2111.04069/) (2021)
- [Refining Latent Representations: A Generative SSL Approach for Heterogeneous Graph Learning](https://mlsanity.com/paper/2310.11102/) (2023)
- [Distortion-Aware Adversarial Attacks on Bounding Boxes of Object Detectors](https://mlsanity.com/paper/2412.18815/) (2024)
- [Visible and infrared self-supervised fusion trained on a single example](https://mlsanity.com/paper/2307.04100/) (2023)

## Abstract (arXiv)

The Mean Square Error (MSE) has shown its strength when applied in deep generative models such as Auto-Encoders to model reconstruction loss. However, in image domain especially, the limitation of MSE is obvious: it assumes pixel independence and ignores spatial relationships of samples. This contradicts most architectures of Auto-Encoders which use convolutional layers to extract spatial dependent features. We base on the structural similarity metric (SSIM) and propose a novel level weighted structural similarity (LWSSIM) loss for convolutional Auto-Encoders. Experiments on common datasets on various Auto-Encoder variants show that our loss is able to outperform the MSE loss and the Vanilla SSIM loss. We also provide reasons why our model is able to succeed in cases where the standard SSIM loss fails.