VIPriors 2: Visual Inductive Priors for Data-Efficient Deep Learning Challenges

TL;DR

This paper presents a set of challenges focused on data-efficient deep learning in computer vision, emphasizing training from scratch with limited data and encouraging innovative inductive bias incorporation.

Contribution

It introduces five new challenges that promote research on data-efficient models without transfer learning, highlighting effective strategies like data augmentation and ensembling.

Findings

Baseline models are significantly outperformed by proposed methods.

Data augmentation and ensembling are key to improving data efficiency.

Models trained from scratch can achieve competitive performance without transfer learning.

Abstract

The second edition of the "VIPriors: Visual Inductive Priors for Data-Efficient Deep Learning" challenges featured five data-impaired challenges, where models are trained from scratch on a reduced number of training samples for various key computer vision tasks. To encourage new and creative ideas on incorporating relevant inductive biases to improve the data efficiency of deep learning models, we prohibited the use of pre-trained checkpoints and other transfer learning techniques. The provided baselines are outperformed by a large margin in all five challenges, mainly thanks to extensive data augmentation policies, model ensembling, and data efficient network architectures.