Abstract:Feature pyramid networks (FPN) are widely exploited for multi-scale feature fusion in existing advanced object detection frameworks. Numerous previous works have developed various structures for bidirectional feature fusion, all of which are shown to improve the detection performance effectively. We observe that these complicated network structures require feature pyramids to be stacked in a fixed order, which introduces longer pipelines and reduces the inference speed. Moreover, semantics from non-adjacent levels are diluted in the feature pyramid since only features at adjacent pyramid levels are merged by the local fusion operation in a sequence manner. To address these issues, we propose a novel architecture named RCNet, which consists of Reverse Feature Pyramid (RevFP) and Cross-scale Shift Network (CSN). RevFP utilizes local bidirectional feature fusion to simplify the bidirectional pyramid inference pipeline. CSN directly propagates representations to both adjacent and non-adjacent levels to enable multi-scale features more correlative. Extensive experiments on the MS COCO dataset demonstrate RCNet can consistently bring significant improvements over both one-stage and two-stage detectors with subtle extra computational overhead. In particular, RetinaNet is boosted to 40.2 AP, which is 3.7 points higher than baseline, by replacing FPN with our proposed model. On COCO test-dev, RCNet can achieve very competitive performance with a single-model single-scale 50.5 AP. Codes will be made available.

Featurized Query R-CNN

Can the Query-based Object Detector Be Designed with Fewer Stages?

QueryDet: Cascaded Sparse Query for Accelerating High-Resolution Small Object Detection

DECO: Query-Based End-to-End Object Detection with ConvNets

Conditional DETR V2: Efficient Detection Transformer with Box Queries

Efficient object detector via dynamic prior and dynamic feature fusion

Complementary Feature Pyramid Network for Object Detection

RCNet: Reverse Feature Pyramid and Cross-scale Shift Network for Object Detection

Dense Distinct Query for End-to-End Object Detection

Anchor DETR: Query Design for Transformer-Based Detector

Team DETR: Guide Queries As a Professional Team in Detection Transformers

Comprehensive Feature Enhancement Module For Single-Shot Object Detector

Feature Combination Based On Receptive Fields And Cross-Fusion Feature Pyramid For Object Detection

VFEDet: a variational information bottleneck based feature enhancement object detection network

CA2Det: Cascaded Adaptive Fusion Pyramid Network Based on Attention Mechanism for Small Object Detection

You Only Look One-level Feature

Adaptive Query Selection for Camouflaged Instance Segmentation

A New Feature Pyramid Network for Object Detection

PMG-DETR: fast convergence of DETR with position-sensitive multi-scale attention and grouped queries

A novel fast combine-and-conquer object detector based on only one-level feature map

DETR++: Taming Your Multi-Scale Detection Transformer