Self-improving object detection via disagreement reconciliation
This addresses the issue of performance drops in object detection for applications like robotics or autonomous systems when deployed in unseen conditions, though it is incremental as it builds on existing self-supervised techniques.
The paper tackles the problem of object detectors losing performance in new environments by proposing a self-supervised method that fine-tunes detectors without human intervention, resulting in a 2.66% mAP improvement over an off-the-shelf detector and outperforming state-of-the-art methods.
Object detectors often experience a drop in performance when new environmental conditions are insufficiently represented in the training data. This paper studies how to automatically fine-tune a pre-existing object detector while exploring and acquiring images in a new environment without relying on human intervention, i.e., in a self-supervised fashion. In our setting, an agent initially explores the environment using a pre-trained off-the-shelf detector to locate objects and associate pseudo-labels. By assuming that pseudo-labels for the same object must be consistent across different views, we devise a novel mechanism for producing refined predictions from the consensus among observations. Our approach improves the off-the-shelf object detector by 2.66% in terms of mAP and outperforms the current state of the art without relying on ground-truth annotations.