Abstract:In the CNN based object detectors, feature pyramids are widely exploited to alleviate the problem of scale variation across object instances. These object detectors, which strengthen features via a top-down pathway and lateral connections, are mainly to enrich the semantic information of low-level features, but ignore the enhancement of high-level features. This can lead to an imbalance between different levels of features, in particular a serious lack of detailed information in the high-level features, which makes it difficult to get accurate bounding boxes. In this paper, we introduce a novel two-pronged transductive idea to explore the relationship among different layers in both backward and forward directions, which can enrich the semantic information of low-level features and detailed information of high-level features at the same time. Under the guidance of the two-pronged idea, we propose a Two-Pronged Network (TPNet) to achieve bidirectional transfer between high-level features and low-level features, which is useful for accurately detecting object at different scales. Furthermore, due to the distribution imbalance between the hard and easy samples in single-stage detectors, the gradient of localization loss is always dominated by the hard examples that have poor localization accuracy. This will enable the model to be biased toward the hard samples. So in our TPNet, an adaptive IoU based localization loss, named Rectified IoU (RIoU) loss, is proposed to rectify the gradients of each kind of samples. The Rectified IoU loss increases the gradients of examples with high IoU while suppressing the gradients of examples with low IoU, which can improve the overall localization accuracy of model. Extensive experiments demonstrate the superiority of our TPNet and RIoU loss.

IoU-uniform R-CNN: Breaking Through the Limitations of RPN

IoU-aware Single-stage Object Detector for Accurate Localization

Research on Low-Resolution Pedestrian Detection Algorithms Based on R-CNN with Targeted Pooling and Proposal

C-RPNs: Promoting object detection in real world via a cascade structure of Region Proposal Networks.

IoU-Adaptive Deformable R-CNN: Make Full Use of IoU for Multi-Class Object Detection in Remote Sensing Imagery

3D IoU-Net: IoU Guided 3D Object Detector for Point Clouds

Unified-IoU: For High-Quality Object Detection

Rethink the IoU-based Loss Functions for Bounding Box Regression

Cascade R-CNN: Delving into High Quality Object Detection

Single-Shot Two-Pronged Detector with Rectified IoU Loss

DeRPN: Taking a Further Step Toward More General Object Detection.

Bal-R$^2$CNN: High Quality Recurrent Object Detection with Balance Optimization

RON: Reverse Connection with Objectness Prior Networks for Object Detection

Rethinking IoU-based Optimization for Single-stage 3D Object Detection

Training Cascade Compact CNN With Region-IoU for Accurate Pedestrian Detection

From Voxel to Point: IoU-guided 3D Object Detection for Point Cloud with Voxel-to-Point Decoder

NIV-SSD: Neighbor IoU-Voting Single-Stage Object Detector From Point Cloud

Revisiting Faster R-Cnn: A Deeper Look At Region Proposal Network

Delving deep into the imbalance of positive proposals in two-stage object detection

Robust Bounding Box Regression for Small Object Detection

Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks