SASSL: Enhancing Self-Supervised Learning via Neural Style Transfer

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Rojas-Gomez, Renan A., Singhal, Karan, Etemad, Ali, Bijamov, Alex, Morningstar, Warren R., Mansfield, Philip Andrew
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915002043072512
author Rojas-Gomez, Renan A.
Singhal, Karan
Etemad, Ali
Bijamov, Alex
Morningstar, Warren R.
Mansfield, Philip Andrew
author_facet Rojas-Gomez, Renan A.
Singhal, Karan
Etemad, Ali
Bijamov, Alex
Morningstar, Warren R.
Mansfield, Philip Andrew
contents Existing data augmentation in self-supervised learning, while diverse, fails to preserve the inherent structure of natural images. This results in distorted augmented samples with compromised semantic information, ultimately impacting downstream performance. To overcome this limitation, we propose SASSL: Style Augmentations for Self Supervised Learning, a novel data augmentation technique based on Neural Style Transfer. SASSL decouples semantic and stylistic attributes in images and applies transformations exclusively to their style while preserving content, generating diverse samples that better retain semantic information. SASSL boosts top-1 image classification accuracy on ImageNet by up to 2 percentage points compared to established self-supervised methods like MoCo, SimCLR, and BYOL, while achieving superior transfer learning performance across various datasets. Because SASSL can be performed asynchronously as part of the data augmentation pipeline, these performance impacts can be obtained with no change in pretraining throughput.
format Preprint
id arxiv_https___arxiv_org_abs_2312_01187
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle SASSL: Enhancing Self-Supervised Learning via Neural Style Transfer
Rojas-Gomez, Renan A.
Singhal, Karan
Etemad, Ali
Bijamov, Alex
Morningstar, Warren R.
Mansfield, Philip Andrew
Computer Vision and Pattern Recognition
Machine Learning
Existing data augmentation in self-supervised learning, while diverse, fails to preserve the inherent structure of natural images. This results in distorted augmented samples with compromised semantic information, ultimately impacting downstream performance. To overcome this limitation, we propose SASSL: Style Augmentations for Self Supervised Learning, a novel data augmentation technique based on Neural Style Transfer. SASSL decouples semantic and stylistic attributes in images and applies transformations exclusively to their style while preserving content, generating diverse samples that better retain semantic information. SASSL boosts top-1 image classification accuracy on ImageNet by up to 2 percentage points compared to established self-supervised methods like MoCo, SimCLR, and BYOL, while achieving superior transfer learning performance across various datasets. Because SASSL can be performed asynchronously as part of the data augmentation pipeline, these performance impacts can be obtained with no change in pretraining throughput.
title SASSL: Enhancing Self-Supervised Learning via Neural Style Transfer
topic Computer Vision and Pattern Recognition
Machine Learning
url https://arxiv.org/abs/2312.01187