Rethinking Features-Fused-Pyramid-Neck for Object Detection

TL;DR

This paper proposes a new architecture and techniques for multi-scale object detection that improve feature alignment and detection accuracy, achieving state-of-the-art results on Pascal VOC and MS COCO datasets.

Contribution

It introduces an independent hierarchy pyramid architecture, soft nearest neighbor interpolation, and adaptive feature selection methods to enhance multi-head detectors.

Findings

Achieved state-of-the-art detection performance on Pascal VOC.

Improved feature alignment with the secondary features alignment solution.

Enhanced lightweight convolutional techniques for real-time detection.

Abstract

Multi-head detectors typically employ a features-fused-pyramid-neck for multi-scale detection and are widely adopted in the industry. However, this approach faces feature misalignment when representations from different hierarchical levels of the feature pyramid are forcibly fused point-to-point. To address this issue, we designed an independent hierarchy pyramid (IHP) architecture to evaluate the effectiveness of the features-unfused-pyramid-neck for multi-head detectors. Subsequently, we introduced soft nearest neighbor interpolation (SNI) with a weight downscaling factor to mitigate the impact of feature fusion at different hierarchies while preserving key textures. Furthermore, we present a features adaptive selection method for down sampling in extended spatial windows (ESD) to retain spatial features and enhance lightweight convolutional techniques (GSConvE). These advancements culminate in our secondary features alignment solution (SA) for real-time detection, achieving state-of-the-art results on Pascal VOC and MS COCO. Code will be released at https://github.com/AlanLi1997/rethinking-fpn. This paper has been accepted by ECCV2024 and published on Springer Nature.