Bounding Box Regression with Uncertainty for Accurate Object Detection

Abstract

Large-scale object detection datasets (e.g., MS-COCO) try to define the ground truth bounding boxes as clear as possible. However, we observe that ambiguities are still introduced when labeling the bounding boxes. In this paper, we propose a novel bounding box regression loss for learning bounding box transformation and localization variance together. Our loss greatly improves the localization accuracies of various architectures with nearly no additional computation. The learned localization variance allows us to merge neighboring bounding boxes during non-maximum suppression (NMS), which further improves the localization performance. On MS-COCO, we boost the Average Precision (AP) of VGG-16 Faster R-CNN from 23.6% to 29.1%. More importantly, for ResNet-50-FPN Mask R-CNN, our method improves the AP and AP90 by 1.8% and 6.2% respectively, which significantly outperforms previous state-of-the-art bounding box refinement methods. Our code and models are available at: github.com/yihui-he/KL-Loss

arXiv commentCVPR 2019

References in corpus (23)

• Very Deep Convolutional Networks for Large-Scale Image Recognition
• Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
• MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications
• R-FCN: Object Detection via Region-based Fully Convolutional Networks
• What Uncertainties Do We Need in Bayesian Deep Learning for Computer Vision?
• Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour
• Going Deeper with Convolutions
• AMC: AutoML for Model Compression and Acceleration on Mobile Devices
• cuDNN: Efficient Primitives for Deep Learning
• Multi-Task Learning Using Uncertainty to Weigh Losses for Scene Geometry and Semantics
• CrowdHuman: A Benchmark for Detecting Human in a Crowd
• Deformable Convolutional Networks
• Cascade R-CNN: Delving into High Quality Object Detection
• Aggregated Residual Transformations for Deep Neural Networks
• Light-Head R-CNN: In Defense of Two-Stage Object Detector
• An Implementation of Faster RCNN with Study for Region Sampling
• Relation Networks for Object Detection
• Uncertainty-aware Short-term Motion Prediction of Traffic Actors for Autonomous Driving
• Vehicle Traffic Driven Camera Placement for Better Metropolis Security Surveillance
• Learning 6-DoF Grasping and Pick-Place Using Attention Focus
• Single Image Super-resolution via a Lightweight Residual Convolutional Neural Network
• Prediction-Tracking-Segmentation
• CIRL: Controllable Imitative Reinforcement Learning for Vision-based Self-driving

Cited by in corpus (7)

• Visibility Guided NMS: Efficient Boosting of Amodal Object Detection in Crowded Traffic Scenes
• Prediction-Tracking-Segmentation
• IoU-uniform R-CNN: Breaking Through the Limitations of RPN
• Depth-wise Decomposition for Accelerating Separable Convolutions in Efficient Convolutional Neural Networks
• iffDetector: Inference-aware Feature Filtering for Object Detection
• Probabilistic Ranking-Aware Ensembles for Enhanced Object Detections
• Detecting Lesion Bounding Ellipses With Gaussian Proposal Networks