Enhancing few-shot object detection through pseudo-label mining

Few-shot object detection involves adapting an existing detector to a set of unseen categories with few annotated examples. This data limitation makes these methods to underperform those trained on large labeled datasets. In many scenarios, there is a high amount of unlabeled data that is never exploited. Thus, we propose to e xPAND the initial novel set by mining pseudo-labels. From a raw set of detections, xPAND obtains reliable pseudo-labels suitable for training any detector. To this end, we propose two new modules: Class and Box confirmation. Class Confirmation aims to remove misclassified pseudo-labels by comparing candidates with expected class prototypes. Box Confirmation estimates IoU to discard inadequately framed objects. Experimental results demonstrate that xPAND enhances the performance of multiple detectors up to +5.9 nAP and +16.4 nAP50 points for MS-COCO and PASCAL VOC, respectively, establishing a new state of the art. Code: https://github.com/PAGF188/xPAND . • We propose xPAND, a mining pipeline that generates high-quality diverse pseudo-labels. • xPAND refines a raw set of pseudo-labels using class and box confirmation modules. • Class module removes misclassified objects by comparing them with expected prototypes. • Box Confirmation estimates IoU to discard inadequately framed objects. • Results show that xPAND sets a new state of the art for both MS-COCO and VOC datasets.