Augmentations vs Algorithms: What Works in Self-Supervised Learning

no code implementations • 8 Mar 2024 • Warren Morningstar, Alex Bijamov, Chris Duvarney, Luke Friedman, Neha Kalibhat, Luyang Liu, Philip Mansfield, Renan Rojas-Gomez, Karan Singhal, Bradley Green, Sushant Prakash

We study the relative effects of data augmentations, pretraining algorithms, and model architectures in Self-Supervised Learning (SSL).

Self-Supervised Learning

SASSL: Enhancing Self-Supervised Learning via Neural Style Transfer

no code implementations • 2 Dec 2023 • Renan A. Rojas-Gomez, Karan Singhal, Ali Etemad, Alex Bijamov, Warren R. Morningstar, Philip Andrew Mansfield

Existing data augmentation in self-supervised learning, while diverse, fails to preserve the inherent structure of natural images.

Data Augmentation Self-Supervised Learning +2

Disentangling the Effects of Data Augmentation and Format Transform in Self-Supervised Learning of Image Representations

no code implementations • 2 Dec 2023 • Neha Kalibhat, Warren Morningstar, Alex Bijamov, Luyang Liu, Karan Singhal, Philip Mansfield

We define augmentations in frequency space called Fourier Domain Augmentations (FDA) and show that training SSL models on a combination of these and image augmentations can improve the downstream classification accuracy by up to 1. 3% on ImageNet-1K.

Data Augmentation Self-Supervised Learning +1