What Makes for Effective Detection Proposals?

@article{Hosang2016WhatMF,
  title={What Makes for Effective Detection Proposals?},
  author={Jan Hendrik Hosang and Rodrigo Benenson and Piotr Doll{\'a}r and Bernt Schiele},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year={2016},
  volume={38},
  pages={814-830}
}
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… 
View With Semantic Reader
623 Citations
Highly Influential Citations
59
Background Citations
209
Methods Citations
213
Results Citations
13

623 Citations

Citation Type

Citation Type

Re-ranking Object Proposals for Object Detection in Automatic Driving
TLDR
A semantic, class-specific approach to re-rank object proposals, which can consistently improve the recall performance even with less proposals, and can achieve high recall rate uner strict critical even using less proposals.
Class-specific object proposals re-ranking for object detection in automatic driving
A comparative study of object proposals re-ranking methods for object detection
TLDR
A comparative study of the existing unsupervised objectness measuring approaches is made to testify their effectiveness and generalization abilities, and demonstrates that contour is not an adequate objectness cue to pop out high quality proposals.
Object proposals using CNN-based edge filtering
TLDR
This paper proposes a novel idea of filtering irrelevant edges using semantic image filtering and true objectness learnt within convolutional layers of CNN, and localizes well proposals by producing highly accurate bounding boxes and reduces the number of proposals.
Boundary-aware box refinement for object proposal generation
Object Proposal Generation for Unsupervised Object Localization
TLDR
The experimental analysis demonstrates that the performance of these methods depends on the number of candidates and the threshold of recall, meanwhile the methods that generate the candidates with scores perform better than those methods that do not with scores.
Object Proposal Generation With Fully Convolutional Networks
TLDR
This paper presents a framework utilizing fully convolutional networks (FCNs) to produce object proposal positions and bounding box location refinement with Support Vector Machine (SVM) to further improve proposal localization.
Robust object proposals re-ranking for object detection in autonomous driving using convolutional neural networks
HyperNet: Towards Accurate Region Proposal Generation and Joint Object Detection
TLDR
This paper presents a deep hierarchical network, namely HyperNet, for handling region proposal generation and object detection jointly, primarily based on an elaborately designed Hyper Feature which aggregates hierarchical feature maps first and then compresses them into a uniform space.
You Reap What You Sow: Using Videos to Generate High Precision Object Proposals for Weakly-Supervised Object Detection
TLDR
This work uses video and motion cues to automatically estimate the extent of objects to train a Weakly-supervised Region Proposal Network (W-RPN), which is used to generate high precision object proposals, which are in turn used to re-rank high recall proposals like edge boxes or selective search according to their spatial overlap.
...
...

References

SHOWING 1-10 OF 81 REFERENCES
How good are detection proposals, really?
TLDR
An in depth analysis of ten object proposal methods along with four baselines regarding ground truth annotation recall (on Pascal VOC 2007 and ImageNet 2013), repeatability, and impact on DPM detector performance are provided.
Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation
TLDR
This paper proposes a simple and scalable detection algorithm that improves mean average precision (mAP) by more than 30% relative to the previous best result on VOC 2012 -- achieving a mAP of 53.3%.
Scalable, High-Quality Object Detection
TLDR
It is demonstrated that learning-based proposal methods can effectively match the performance of hand-engineered methods while allowing for very efficient runtime-quality trade-offs.
Learning to Segment Object Candidates
TLDR
A new way to generate object proposals is proposed, introducing an approach based on a discriminative convolutional network that obtains substantially higher object recall using fewer proposals and is able to generalize to unseen categories it has not seen during training.
Improving Spatial Support for Objects via Multiple Segmentations
TLDR
The multiple segmentation approach is used to evaluate how close can real segments approach the ground-truth for real objects, and at what cost.
Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
TLDR
This work introduces a Region Proposal Network (RPN) that shares full-image convolutional features with the detection network, thus enabling nearly cost-free region proposals and further merge RPN and Fast R-CNN into a single network by sharing their convolutionAL features.
Edge Boxes: Locating Object Proposals from Edges
TLDR
A novel method for generating object bounding box proposals using edges is proposed, showing results that are significantly more accurate than the current state-of-the-art while being faster to compute.
Segmentation as selective search for object recognition
TLDR
This work adapt segmentation as a selective search by reconsidering segmentation to generate many approximate locations over few and precise object delineations because an object whose location is never generated can not be recognised and appearance and immediate nearby context are most effective for object recognition.
Regionlets for Generic Object Detection
TLDR
This work proposes to model an object class by a cascaded boosting classifier which integrates various types of features from competing local regions, named as region lets, which significantly outperforms the state-of-the-art on popular multi-class detection benchmark datasets with a single method.
Measuring the Objectness of Image Windows
TLDR
A generic objectness measure, quantifying how likely it is for an image window to contain an object of any class, and uses objectness as a complementary score in addition to the class-specific model, which leads to fewer false positives.
...
...