What makes for effective detection proposals?

Abstract

Current top performing object detectors employ detection proposals to guide the search for objects, thereby avoiding exhaustive sliding window search across images. Despite the popularity and widespread use of detection proposals, it is unclear which trade-offs are made when using them during object detection. We provide an in-depth analysis of twelve proposal methods along with four baselines regarding proposal repeatability, ground truth annotation recall on PASCAL, ImageNet, and MS COCO, and their impact on DPM, R-CNN, and Fast R-CNN detection performance. Our analysis shows that for object detection improving proposal localisation accuracy is as important as improving recall. We introduce a novel metric, the average recall (AR), which rewards both high recall and good localisation and correlates surprisingly well with detection performance. Our findings show common strengths and weaknesses of existing methods, and provide insights and metrics for selecting and tuning proposal methods.

TPAMI final version, duplicate proposals removed in experiments

References in corpus (2)

• Learning to Segment Object Candidates
• Fast R-CNN

Cited by in corpus (79)

• Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
• Learning to Segment Object Candidates
• Multiscale Combinatorial Grouping for Image Segmentation and Object Proposal Generation
• Object Detection in 20 Years: A Survey
• 3D Object Detection for Autonomous Driving: A Survey
• The EuroCity Persons Dataset: A Novel Benchmark for Object Detection
• Deep Learning for Generic Object Detection: A Survey
• Semantic Instance Segmentation via Deep Metric Learning
• Few-Example Object Detection with Model Communication
• PointRCNN: 3D Object Proposal Generation and Detection from Point Cloud
• From Handcrafted to Deep Features for Pedestrian Detection: A Survey
• Convolutional Oriented Boundaries
• Postdisaster image-based damage detection and repair cost estimation of reinforced concrete buildings using dual convolutional neural networks
• A Novel CNN-based Method for Accurate Ship Detection in HR Optical Remote Sensing Images via Rotated Bounding Box
• Multiclass Weighted Loss for Instance Segmentation of Cluttered Cells
• Tree-Structured Reinforcement Learning for Sequential Object Localization
• Deep-Learning Convolutional Neural Networks for scattered shrub detection with Google Earth Imagery
• Multi-View 3D Object Detection Network for Autonomous Driving
• Inside-Outside Net: Detecting Objects in Context with Skip Pooling and Recurrent Neural Networks
• Oriented object detection in optical remote sensing images using deep learning: a survey
• 3D Object Proposals using Stereo Imagery for Accurate Object Class Detection
• Learning Sampling Distributions for Efficient Object Detection
• A Comprehensive Review of Modern Object Segmentation Approaches
• Where are the Masks: Instance Segmentation with Image-level Supervision
• Simple Does It: Weakly Supervised Instance and Semantic Segmentation
• DeepBox: Learning Objectness with Convolutional Networks
• Deep Neural Networks for Marine Debris Detection in Sonar Images
• Locating 3D Object Proposals: A Depth-Based Online Approach
• Zoom Out-and-In Network with Recursive Training for Object Proposal
• Objectness Scoring and Detection Proposals in Forward-Looking Sonar Images with Convolutional Neural Networks
• Scale-aware Pixel-wise Object Proposal Networks
• Cell Tracking via Proposal Generation and Selection
• Crafting GBD-Net for Object Detection
• Learning to Group: A Bottom-Up Framework for 3D Part Discovery in Unseen Categories
• Boosting Convolutional Features for Robust Object Proposals
• Co-localization with Category-Consistent Features and Geodesic Distance Propagation
• DeepProposal: Hunting Objects by Cascading Deep Convolutional Layers
• Effective Use of Dilated Convolutions for Segmenting Small Object Instances in Remote Sensing Imagery
• Deep Extreme Cut: From Extreme Points to Object Segmentation
• ScaleNet: Guiding Object Proposal Generation in Supermarkets and Beyond
• Context Proposals for Saliency Detection
• Detect2Rank : Combining Object Detectors Using Learning to Rank
• Tracking objects using 3D object proposals
• Finding Deviated Behaviors of the Compressed DNN Models for Image Classifications
• Semantic Amodal Segmentation
• Deep Multiple Instance Learning for Airplane Detection in High Resolution Imagery
• Deep Functional Dictionaries: Learning Consistent Semantic Structures on 3D Models from Functions
• Zoom Out-and-In Network with Map Attention Decision for Region Proposal and Object Detection
• Weakly Supervised Object Boundaries
• Fast Region Proposal Learning for Object Detection for Robotics
• Testing Deep Learning Models for Image Analysis Using Object-Relevant Metamorphic Relations
• Convolutional Oriented Boundaries: From Image Segmentation to High-Level Tasks
• Saliency-guided Adaptive Seeding for Supervoxel Segmentation
• Fast Efficient Object Detection Using Selective Attention
• Motorcycle detection and classification in urban Scenarios using a model based on Faster R-CNN
• Deeply Activated Salient Region for Instance Search
• A Fully Spiking Hybrid Neural Network for Energy-Efficient Object Detection
• UESegNet: Context Aware Unconstrained ROI Segmentation Networks for Ear Biometric
• Object Proposal with Kernelized Partial Ranking
• FashionFail: Addressing Failure Cases in Fashion Object Detection and Segmentation
• Learning from Counting: Leveraging Temporal Classification for Weakly Supervised Object Localization and Detection
• What leads to generalization of object proposals?
• Toward Filament Segmentation Using Deep Neural Networks
• Identify the cells' nuclei based on the deep learning neural network
• The Image Torque Operator for Contour Processing
• Bandwidth limited object recognition in high resolution imagery
• A Study on Evaluation Standard for Automatic Crack Detection Regard the Random Fractal
• Detecting Temporally Consistent Objects in Videos through Object Class Label Propagation
• Robust Anomaly-Based Ship Proposals Detection from Pan-sharpened High-Resolution Satellite Image
• Rethinking Task and Metrics of Instance Segmentation on 3D Point Clouds
• Tubelets: Unsupervised action proposals from spatiotemporal super-voxels
• Cascaded Sparse Spatial Bins for Efficient and Effective Generic Object Detection
• Collaborative Deep Reinforcement Learning for Joint Object Search
• Learning to Segment Object Candidates via Recursive Neural Networks
• Segmentation Free Object Discovery in Video
• Saliency Detection for Improving Object Proposals
• A High-Performance Object Proposals based on Horizontal High Frequency Signal
• Supervision Levels Scale (SLS)
• Character Proposal Network for Robust Text Extraction