How good are detection proposals, really?

By Jan Hendrik Hosang, Rodrigo Benenson, Bernt Schiele. 2014

274 citations recorded by the source corpus.

Current top performing Pascal VOC object detectors employ detection proposals to guide the search for objects thereby avoiding exhaustive sliding window search across images. Despite the popularity of detection proposals, it is unclear which trade-offs are made when using them during object detection. We provide 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. Our findings show common weaknesses of existing methods, and provide insights to choose the most adequate method for different settings.

Subfields: Computer Vision and Pattern Recognition, Evaluation & Benchmarks

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Evaluation & Benchmarks

Citations in the corpus

Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks (2015, methodology): “In the following, we train Fast R-CNN using 2k RPN proposals, but evaluate different numbers of proposals at test-time.”
Scalable, High-Quality Object Detection (2014, neutral): “We emphasize that MultiBox and SPP are complementary in the sense that spatial pyramid pooling can be added to the underlying ConvNet if desired, and post-classification of proposals can be sped up in the same way with no change to the MultiBox objective.”
Attend Refine Repeat: Active Box Proposal Generation via In-Out Localization (2016, background): “[21, 22] where in their work they demonstrated that it correlates well with the average precision performance of box proposal based object detection systems.”
CRAFT Objects from Images (2016, background): “Many works have been proposed for further improvement and an evaluation can be found in [16].”
Self-taught object localization with deep networks (2014, methodology): “Objectness methods [2, 6, 30, 4, 3, 16, 36, 25, 21, 26, 32, 14] aim at generating bounding boxes that yield high recall for a high number of candidates, i.”
Object proposal by multi-branch hierarchical segmentation (2015, background): “A more comprehensive survey is provided in [18].”
ME R-CNN: Multi-Expert Region-based CNN for Object Detection (2017, background): “Several more approaches [5, 6, 18, 19, 33, 37] to generate RoIs based on objectness characteristics have been introduced afterwards.”
Relief Impression Image Detection : Unsupervised Extracting Objects Directly From Feature Arrangements of Deep CNN (2016, methodology): “In section V-C we use the region proposals generated by [17] and the evaluation code deployed by [33], but this is just for the purpose of saving time, because there will be no difference if we run these methods again on our machine.”
Image‐based post‐disaster inspection of reinforced concrete bridge systems using deep learning with Bayesian optimization (2018, methodology): “Comprehensive surveys and comparisons of object proposal methods can be found in Hosang, Benenson, and Schiele (2014), Hosang, Benenson, Dollár, and Schiele (2015), and Chavali, Agrawal, Mahendru, and Batra (2015).”
Beyond pixels: A comprehensive survey from bottom-up to semantic image segmentation and cosegmentation (2015, background): “A more complete survey of state-of-the-art object-ness window methods and salient object detection methods can be found in [95] and [96], respectively.”
MultiNet: Real-time Joint Semantic Reasoning for Autonomous Driving (2016, methodology): “Detection: Traditional deep learning approaches to object detection follow a two step process, where region proposals [25, 20, 19] are ﬁrst generated and then scored using a convolutional network [15, 40].”
Accurate Object Localization in Remote Sensing Images Based on Convolutional Neural Networks (2017, background): “[48] provided a deep analysis concerning ten different object-proposal methods and found that the recall of the selective search method is approximately 0.”
Edge Boxes: Locating Object Proposals from Edges (2014, background): “[26] for a thorough survey and evaluation of object proposal methods.”
Cross-modal adaptation for RGB-D detection (2016, background): “Given the importance of good region proposals [25], naturally people have studied the problem of using depth images to improve the quality of object proposals [34], [20].”