Generalized Focal Loss V2: Learning Reliable Localization Quality Estimation for Dense Object Detection (2020-11-25T00:00:00.000000Z)

TL;DR

This paper explores a completely novel and different perspective to perform LQE – based on the learned distributions of the four parameters of the bounding box – and develops a considerably lightweight Distribution-Guided Quality Predictor (DGQP) for reliable LqE based on GFLV1, thus producing GFLv2.

Abstract

Localization Quality Estimation (LQE) is crucial and popular in the recent advancement of dense object detectors since it can provide accurate ranking scores that benefit the Non-Maximum Suppression processing and improve detection performance. As a common practice, most existing methods predict LQE scores through vanilla convolutional features shared with object classification or bounding box regression. In this paper, we explore a completely novel and different perspective to perform LQE – based on the learned distributions of the four parameters of the bounding box. The bounding box distributions are inspired and introduced as "General Distribution" in GFLV1, which describes the uncertainty of the predicted bounding boxes well. Such a property makes the distribution statistics of a bounding box highly correlated to its real localization quality. Specifically, a bounding box distribution with a sharp peak usually corresponds to high localization quality, and vice versa. By leveraging the close correlation between distribution statistics and the real localization quality, we develop a considerably lightweight Distribution-Guided Quality Predictor (DGQP) for reliable LQE based on GFLV1, thus producing GFLV2. To our best knowledge, it is the first attempt in object detection to use a highly relevant, statistical representation to facilitate LQE. Extensive experiments demonstrate the effectiveness of our method. Notably, GFLV2 (ResNet101) achieves 46.2 AP at 14.6 FPS, surpassing the previous state-of-the-art ATSS baseline (43.6 AP at 14.6 FPS) by absolute 2.6 AP on COCO test-dev, without sacrificing the efficiency both in training and inference.

Authors

References46 items

Learning to Match Anchors for Visual Object Detection

VarifocalNet: An IoU-aware Dense Object Detector

BorderDet: Border Feature for Dense Object Detection

Probabilistic Anchor Assignment with IoU Prediction for Object Detection

RepPoints V2: Verification Meets Regression for Object Detection

AutoAssign: Differentiable Label Assignment for Dense Object Detection

Generalized Focal Loss: Learning Qualified and Distributed Bounding Boxes for Dense Object Detection

CentripetalNet: Pursuing High-Quality Keypoint Pairs for Object Detection

Revisiting the Sibling Head in Object Detector

IoU-aware Single-stage Object Detector for Accurate Localization

Side-Aware Boundary Localization for More Precise Object Detection

Bridging the Gap Between Anchor-Based and Anchor-Free Detection via Adaptive Training Sample Selection

Soft Anchor-Point Object Detection

Hierarchical Shot Detector

MMDetection: Open MMLab Detection Toolbox and Benchmark

RepPoints: Point Set Representation for Object Detection

CenterNet: Keypoint Triplets for Object Detection

Objects as Points

FoveaBox: Beyond Anchor-based Object Detector

Libra R-CNN: Towards Balanced Learning for Object Detection

Res2Net: A New Multi-Scale Backbone Architecture

FCOS: Fully Convolutional One-Stage Object Detection

Feature Selective Anchor-Free Module for Single-Shot Object Detection

Mask Scoring R-CNN

Scale-Aware Trident Networks for Object Detection

Grid R-CNN

Deformable ConvNets V2: More Deformable, Better Results

CornerNet: Detecting Objects as Paired Keypoints

Acquisition of Localization Confidence for Accurate Object Detection

YOLOv3: An Incremental Improvement

Cascade R-CNN: Delving Into High Quality Object Detection

Improving Object Localization with Fitness NMS and Bounded IoU Loss

Focal Loss for Dense Object Detection

Mask R-CNN

Deformable Convolutional Networks

YOLO9000: Better, Faster, Stronger

Feature Pyramid Networks for Object Detection

Aggregated Residual Transformations for Deep Neural Networks

Stacked Hourglass Networks for Human Pose Estimation

Deep Residual Learning for Image Recognition

SSD: Single Shot MultiBox Detector

You Only Look Once: Unified, Real-Time Object Detection

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

Fast R-CNN

Microsoft COCO: Common Objects in Context

ImageNet classification with deep convolutional neural networks