# Data is Overrated: Perceptual Metrics Can Lead Learning in the Absence of Training Data

arXiv: https://arxiv.org/abs/2312.03455 · HTML: https://mlsanity.com/paper/2312.03455/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 5 of its 16 references that are in the map.
- First posted on arXiv 6 Dec 2023.

## Builds on (most cited first, 5 of 5)

- [Generative Adversarial Networks for Extreme Learned Image Compression](https://mlsanity.com/paper/1804.02958/) (2018)
- [Evaluating generative audio systems and their metrics](https://mlsanity.com/paper/2209.00130/) (2022)
- [On the relation between statistical learning and perceptual distances](https://mlsanity.com/paper/2106.04427/) (2021)
- [Enforcing Perceptual Consistency on Generative Adversarial Networks by Using the Normalised Laplacian Pyramid Distance](https://mlsanity.com/paper/1908.04347/) (2019)
- [Perceiving Music Quality with GANs](https://mlsanity.com/paper/2006.06287/) (2020)

## Built on by (most cited first, 1 of 1)

- [The Effect of Perceptual Metrics on Music Representation Learning for Genre Classification](https://mlsanity.com/paper/2409.17069/) (2024)

## Abstract (arXiv)

Perceptual metrics are traditionally used to evaluate the quality of natural signals, such as images and audio. They are designed to mimic the perceptual behaviour of human observers and usually reflect structures found in natural signals. This motivates their use as loss functions for training generative models such that models will learn to capture the structure held in the metric. We take this idea to the extreme in the audio domain by training a compressive autoencoder to reconstruct uniform noise, in lieu of natural data. We show that training with perceptual losses improves the reconstruction of spectrograms and re-synthesized audio at test time over models trained with a standard Euclidean loss. This demonstrates better generalisation to unseen natural signals when using perceptual metrics.