Aligning Pretraining for Detection via Object-Level Contrastive Learning
This work addresses the problem of improving object detection performance for computer vision researchers by introducing a task-aligned pretraining approach, though it is incremental as it builds on existing contrastive learning and detection frameworks.
The paper tackles the sub-optimal transfer of image-level contrastive pretraining to object detection by proposing SoCo, a method that aligns pretraining with detection through object-level representations and architectural modules, achieving state-of-the-art results on COCO detection with Mask R-CNN.
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.