Aligning Pretraining for Detection via Object-Level Contrastive Learning

TL;DR

This paper introduces SoCo, a pretraining method that aligns self-supervised learning with object detection tasks by focusing on object-level representations and detection-specific properties, leading to improved transfer performance.

Contribution

The paper proposes a novel object-level contrastive pretraining method, SoCo, that aligns pretraining with detection tasks by incorporating object proposals and detection modules.

Findings

Achieves state-of-the-art transfer results on COCO detection.

Outperforms previous methods in object detection transfer tasks.

Demonstrates the effectiveness of object-level alignment in pretraining.

Abstract

Image-level contrastive representation learning has proven to be highly effective as a generic model for transfer learning. Such generality for transfer learning, however, sacrifices specificity if we are interested in a certain downstream task. We argue that this could be sub-optimal and thus advocate a design principle which encourages alignment between the self-supervised pretext task and the downstream task. In this paper, we follow this principle with a pretraining method specifically designed for the task of object detection. We attain alignment in the following three aspects: 1) object-level representations are introduced via selective search bounding boxes as object proposals; 2) the pretraining network architecture incorporates the same dedicated modules used in the detection pipeline (e.g. FPN); 3) the pretraining is equipped with object detection properties such as object-level translation invariance and scale invariance. Our method, called Selective Object COntrastive learning (SoCo), achieves state-of-the-art results for transfer performance on COCO detection using a Mask R-CNN framework. Code is available at https://github.com/hologerry/SoCo.