Guided Attention in CNNs for Occluded Pedestrian Detection and Re-identification

Pedestrian detection and re-identification have progressed significantly in the last few years. However, occluded people are notoriously hard to detect and recognize, as their appearance varies substantially depending on a wide range of occlusion patterns. In this paper, we aim to propose a simple and compact method based on CNNs for occlusion handling. We start with interpreting CNN channel features of a pedestrian detector, and we find that different channels activate responses for different body parts respectively. These findings motivate us to employ an attention mechanism across channels to represent various occlusion patterns in one single model, as each occlusion pattern can be formulated as some specific combination of body parts. Therefore, an attention network with self or external guidances is proposed as an add-on to the baseline CNN method. Also, we propose an attention guided self-paced learning method to balance the optimization across different occlusion levels. Our proposed method shows significant improvements over the baseline methods for both pedestrian detection and re-identification tasks. For pedestrian detection, we achieve a considerable improvement of 8pp to the baseline FasterRCNN detector on the heavy occlusion subset of CityPersons and on Caltech we outperform the state-of-the-art method by 5pp. For pedestrian re-identification, our method surpasses the baseline and achieves state-of-the-art performance on multiple re-identification benchmarks.