Multi-Grained Contrast for Data-Efficient Unsupervised Representation Learning

TL;DR

This paper introduces a multi-grained contrastive learning method that enhances unsupervised image representation by capturing multiple levels of detail, leading to better transferability and performance across various downstream tasks without large-scale pretraining.

Contribution

The paper proposes a novel Multi-Grained Contrast (MGC) method that learns representations at multiple granularity levels, improving transferability and performance in unsupervised learning.

Findings

Outperforms state-of-the-art methods on multiple downstream tasks

Demonstrates data-efficient learning without large-scale pretraining

Shows strong transferability of learned representations

Abstract

The existing contrastive learning methods mainly focus on single-grained representation learning, e.g., part-level, object-level or scene-level ones, thus inevitably neglecting the transferability of representations on other granularity levels. In this paper, we aim to learn multi-grained representations, which can effectively describe the image on various granularity levels, thus improving generalization on extensive downstream tasks. To this end, we propose a novel Multi-Grained Contrast method (MGC) for unsupervised representation learning. Specifically, we construct delicate multi-grained correspondences between positive views and then conduct multi-grained contrast by the correspondences to learn more general unsupervised representations. Without pretrained on large-scale dataset, our method significantly outperforms the existing state-of-the-art methods on extensive downstream tasks, including object detection, instance segmentation, scene parsing, semantic segmentation and keypoint detection. Moreover, experimental results support the data-efficient property and excellent representation transferability of our method. The source code and trained weights are available at \url{https://github.com/visresearch/mgc}.