Representation Learning by Learning to Count

TL;DR

This paper presents a novel representation learning method that leverages counting visual primitives through equivariance relations, eliminating manual annotations and improving transfer learning performance.

Contribution

It introduces a counting-based supervision signal derived from image transformations, enabling unsupervised representation learning without manual labels.

Findings

Achieves state-of-the-art transfer learning results

Uses scale and tiling transformations for supervision

Demonstrates effectiveness of counting-based supervision

Abstract

We introduce a novel method for representation learning that uses an artificial supervision signal based on counting visual primitives. This supervision signal is obtained from an equivariance relation, which does not require any manual annotation. We relate transformations of images to transformations of the representations. More specifically, we look for the representation that satisfies such relation rather than the transformations that match a given representation. In this paper, we use two image transformations in the context of counting: scaling and tiling. The first transformation exploits the fact that the number of visual primitives should be invariant to scale. The second transformation allows us to equate the total number of visual primitives in each tile to that in the whole image. These two transformations are combined in one constraint and used to train a neural network with a contrastive loss. The proposed task produces representations that perform on par or exceed the state of the art in transfer learning benchmarks.