Computer Science > Computer Vision and Pattern Recognition
[Submitted on 14 Nov 2015 (this version), latest version 22 May 2017 (v3)]
Title:BING++: A Fast High Quality Object Proposal Generator at 100fps
View PDFAbstract:We are motivated by the need for an object proposal generation algorithm that achieves a good balance between proposal localization quality, object recall and computational efficiency. We propose a novel object proposal algorithm {\em BING++} which inherits the good computational efficiency of BING \cite{BingObj2014} but significantly improves its proposal localization quality. Central to our success is based on the observation that good bounding boxes are those that tightly cover objects. Edge features, which can be computed efficiently, play a critical role in this context. We propose a new algorithm that recursively improves BING's proposals by exploiting the fact that edges in images are typically associated with object boundaries. BING++ improves proposals recursively by incorporating nearest edge points (to proposal boundary pixels) to obtain a tighter bounding box. This operation has linear computational complexity in number of pixels and can be done efficiently using distance transform. Superpixel merging techniques are then employed as post-processing to further improve the proposal quality. Empirically on the VOC2007 dataset, using $10^3$ proposals and IoU threshold 0.5, our method achieves 95.3\% object detection recall (DR), 79.2\% mean average best overlap (MABO), and 68.7\% mean average precision (mAP) on object detection over 20 object classes within an average time of {\bf 0.009} seconds per image.
Submission history
From: Ziming Zhang [view email][v1] Sat, 14 Nov 2015 05:45:47 UTC (9,712 KB)
[v2] Thu, 28 Jul 2016 04:35:15 UTC (3,911 KB)
[v3] Mon, 22 May 2017 17:23:07 UTC (6,824 KB)
Bibliographic and Citation Tools
Bibliographic Explorer (What is the Explorer?)
Litmaps (What is Litmaps?)
scite Smart Citations (What are Smart Citations?)
Code, Data and Media Associated with this Article
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub (What is DagsHub?)
Gotit.pub (What is GotitPub?)
Papers with Code (What is Papers with Code?)
ScienceCast (What is ScienceCast?)
Demos
Recommenders and Search Tools
Influence Flower (What are Influence Flowers?)
Connected Papers (What is Connected Papers?)
CORE Recommender (What is CORE?)
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.