UnitBox: An Advanced Object Detection Network

Abstract

In present object detection systems, the deep convolutional neural networks (CNNs) are utilized to predict bounding boxes of object candidates, and have gained performance advantages over the traditional region proposal methods. However, existing deep CNN methods assume the object bounds to be four independent variables, which could be regressed by the ${\ell }_2$ loss separately. Such an oversimplified assumption is contrary to the well-received observation, that those variables are correlated, resulting to less accurate localization. To address the issue, we firstly introduce a novel Intersection over Union ($IoU$) loss function for bounding box prediction, which regresses the four bounds of a predicted box as a whole unit. By taking the advantages of $IoU$ loss and deep fully convolutional networks, the UnitBox is introduced, which performs accurate and efficient localization, shows robust to objects of varied shapes and scales, and converges fast. We apply UnitBox on face detection task and achieve the best performance among all published methods on the FDDB benchmark.

To appear in ACM MM 2016, 5 pages, 6 figures

References in corpus (4)

• Very Deep Convolutional Networks for Large-Scale Image Recognition
• Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
• DenseBox: Unifying Landmark Localization with End to End Object Detection
• Studying Very Low Resolution Recognition Using Deep Networks

Cited by in corpus (43)

• A Survey of Deep Learning-based Object Detection
• FoveaBox: Beyond Anchor-based Object Detector
• Slim-neck by GSConv: A lightweight-design for real-time detector architectures
• Counting of Grapevine Berries in Images via Semantic Segmentation using Convolutional Neural Networks
• EAST: An Efficient and Accurate Scene Text Detector
• G-Rep: Gaussian Representation for Arbitrary-Oriented Object Detection
• Gaussian Bounding Boxes and Probabilistic Intersection-over-Union for Object Detection
• Towards End-to-End Face Recognition through Alignment Learning
• A Systematic IoU-Related Method: Beyond Simplified Regression for Better Localization
• RDP-Net: Region Detail Preserving Network for Change Detection
• AIParsing: Anchor-free Instance-level Human Parsing
• Regularized Densely-connected Pyramid Network for Salient Instance Segmentation
• Anchor Cascade for Efficient Face Detection
• Dual Semantic Fusion Network for Video Object Detection
• FoodLogoDet-1500: A Dataset for Large-Scale Food Logo Detection via Multi-Scale Feature Decoupling Network
• Single-Shot Two-Pronged Detector with Rectified IoU Loss
• RefineDetLite: A Lightweight One-stage Object Detection Framework for CPU-only Devices
• TDIOT: Target-driven Inference for Deep Video Object Tracking
• SFA: Small Faces Attention Face Detector
• Towards Interpretable and Robust Hand Detection via Pixel-wise Prediction
• Scanning Only Once: An End-to-end Framework for Fast Temporal Grounding in Long Videos
• Learning Category- and Instance-Aware Pixel Embedding for Fast Panoptic Segmentation
• Implicit Motion-Compensated Network for Unsupervised Video Object Segmentation
• Gaussian Combined Distance: A Generic Metric for Object Detection
• Narrowing the Gap: Improved Detector Training with Noisy Location Annotations
• Rotated Object Detection via Scale-invariant Mahalanobis Distance in Aerial Images
• Table Detection for Visually Rich Document Images
• A Lightweight NMS-free Framework for Real-time Visual Fault Detection System of Freight Trains
• FOTS: Fast Oriented Text Spotting with a Unified Network
• SSF-Net: Spatial-Spectral Fusion Network with Spectral Angle Awareness for Hyperspectral Object Tracking
• A Driver Fatigue Recognition Algorithm Based on Spatio-Temporal Feature Sequence
• Anchor Retouching via Model Interaction for Robust Object Detection in Aerial Images
• Road detection via a dual-task network based on cross-layer graph fusion modules
• IVCR-200K: A Large-Scale Multi-turn Dialogue Benchmark for Interactive Video Corpus Retrieval
• Bounding-box deep calibration for high performance face detection
• Box2Poly: Memory-Efficient Polygon Prediction of Arbitrarily Shaped and Rotated Text
• UESegNet: Context Aware Unconstrained ROI Segmentation Networks for Ear Biometric
• Hard Negative Samples Emphasis Tracker without Anchors
• KPNet: Towards Minimal Face Detector
• EC-IoU: Orienting Safety for Object Detectors via Ego-Centric Intersection-over-Union
• Staged Depthwise Correlation and Feature Fusion for Siamese Object Tracking
• Orthogonal Activation with Implicit Group-Aware Bias Learning for Class Imbalance
• MeronymNet: A Hierarchical Approach for Unified and Controllable Multi-Category Object Generation