Deep Regionlets: Blended Representation and Deep Learning for Generic Object Detection

In this article, we propose a novel object detection algorithm named ”Deep Regionlets” by integrating deep neural networks and conventional schema for accurate generic detection. Motivated the effectiveness of regionlets modeling deformations multiple aspect ratios, incorporate into an end-to-end trainable learning framework. The framework consists region selection network regionlet module. Specifically, given bounding box proposal, provides guidance on where to select sub-regions from which features can be learned from. An proposal typically contains three – 16 sub-regions. module focuses local feature transformations alleviate effects appearance variations. To end, first realize non-rectangular within accommodate variations in appearance. Moreover, design “gating network” leaning enable instance dependent soft pooling. Deep Regionlets is trained without additional efforts. We present ablation studies extensive experiments PASCAL VOC dataset Microsoft COCO dataset. proposed method yields competitive performance over state-of-the-art algorithms, such as RetinaNet Mask R-CNN, even segmentation labels.