30Hz object detection with DPM V5

Mohammad Amin Sadeghi, David Forsyth

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

We describe an implementation of the Deformable Parts Model [1] that operates in a user-defined time-frame. Our implementation uses a variety of mechanism to trade-off speed against accuracy. Our implementation can detect all 20 PASCAL 2007 objects simultaneously at 30Hz with an mAP of 0.26. At 15Hz, its mAP is 0.30; and at 100Hz, its mAP is 0.16. By comparison the reference implementation of [1] runs at 0.07Hz and mAP of 0.33 and a fast GPU implementation runs at 1Hz. Our technique is over an order of magnitude faster than the previous fastest DPM implementation. Our implementation exploits a series of important speedup mechanisms. We use the cascade framework of [3] and the vector quantization technique of [2]. To speed up feature computation, we compute HOG features at few scales, and apply many interpolated templates. A hierarchical vector quantization method is used to compress HOG features for fast template evaluation. An object proposal step uses hash-table methods to identify locations where evaluating templates would be most useful; these locations are inserted into a priority queue, and processed in a detection phase. Both proposal and detection phases have an any-time property. Our method applies to legacy templates, and no retraining is required.

Original languageEnglish (US)
Title of host publicationComputer Vision, ECCV 2014 - 13th European Conference, Proceedings
PublisherSpringer-Verlag
Pages65-79
Number of pages15
EditionPART 1
ISBN (Print)9783319105895
DOIs
StatePublished - Jan 1 2014
Event13th European Conference on Computer Vision, ECCV 2014 - Zurich, Switzerland
Duration: Sep 6 2014Sep 12 2014

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
NumberPART 1
Volume8689 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Other

Other13th European Conference on Computer Vision, ECCV 2014
CountrySwitzerland
CityZurich
Period9/6/149/12/14

Fingerprint

Object Detection
Vector quantization
Template
Vector Quantization
Speedup
Priority Queue
Object detection
Cascade
Table
Trade-offs
Series
Evaluation
Graphics processing unit

Keywords

  • Fast Deformable Parts Model
  • Fast Object Detection
  • Real-time Object Detection

ASJC Scopus subject areas

  • Theoretical Computer Science
  • Computer Science(all)

Cite this

Sadeghi, M. A., & Forsyth, D. (2014). 30Hz object detection with DPM V5. In Computer Vision, ECCV 2014 - 13th European Conference, Proceedings (PART 1 ed., pp. 65-79). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 8689 LNCS, No. PART 1). Springer-Verlag. https://doi.org/10.1007/978-3-319-10590-1_5

30Hz object detection with DPM V5. / Sadeghi, Mohammad Amin; Forsyth, David.

Computer Vision, ECCV 2014 - 13th European Conference, Proceedings. PART 1. ed. Springer-Verlag, 2014. p. 65-79 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 8689 LNCS, No. PART 1).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Sadeghi, MA & Forsyth, D 2014, 30Hz object detection with DPM V5. in Computer Vision, ECCV 2014 - 13th European Conference, Proceedings. PART 1 edn, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), no. PART 1, vol. 8689 LNCS, Springer-Verlag, pp. 65-79, 13th European Conference on Computer Vision, ECCV 2014, Zurich, Switzerland, 9/6/14. https://doi.org/10.1007/978-3-319-10590-1_5
Sadeghi MA, Forsyth D. 30Hz object detection with DPM V5. In Computer Vision, ECCV 2014 - 13th European Conference, Proceedings. PART 1 ed. Springer-Verlag. 2014. p. 65-79. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); PART 1). https://doi.org/10.1007/978-3-319-10590-1_5
Sadeghi, Mohammad Amin ; Forsyth, David. / 30Hz object detection with DPM V5. Computer Vision, ECCV 2014 - 13th European Conference, Proceedings. PART 1. ed. Springer-Verlag, 2014. pp. 65-79 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); PART 1).
@inproceedings{f64cbc0546db4b58b8da88dfc1cb15f3,
title = "30Hz object detection with DPM V5",
abstract = "We describe an implementation of the Deformable Parts Model [1] that operates in a user-defined time-frame. Our implementation uses a variety of mechanism to trade-off speed against accuracy. Our implementation can detect all 20 PASCAL 2007 objects simultaneously at 30Hz with an mAP of 0.26. At 15Hz, its mAP is 0.30; and at 100Hz, its mAP is 0.16. By comparison the reference implementation of [1] runs at 0.07Hz and mAP of 0.33 and a fast GPU implementation runs at 1Hz. Our technique is over an order of magnitude faster than the previous fastest DPM implementation. Our implementation exploits a series of important speedup mechanisms. We use the cascade framework of [3] and the vector quantization technique of [2]. To speed up feature computation, we compute HOG features at few scales, and apply many interpolated templates. A hierarchical vector quantization method is used to compress HOG features for fast template evaluation. An object proposal step uses hash-table methods to identify locations where evaluating templates would be most useful; these locations are inserted into a priority queue, and processed in a detection phase. Both proposal and detection phases have an any-time property. Our method applies to legacy templates, and no retraining is required.",
keywords = "Fast Deformable Parts Model, Fast Object Detection, Real-time Object Detection",
author = "Sadeghi, {Mohammad Amin} and David Forsyth",
year = "2014",
month = "1",
day = "1",
doi = "10.1007/978-3-319-10590-1_5",
language = "English (US)",
isbn = "9783319105895",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer-Verlag",
number = "PART 1",
pages = "65--79",
booktitle = "Computer Vision, ECCV 2014 - 13th European Conference, Proceedings",
edition = "PART 1",

}

TY - GEN

T1 - 30Hz object detection with DPM V5

AU - Sadeghi, Mohammad Amin

AU - Forsyth, David

PY - 2014/1/1

Y1 - 2014/1/1

N2 - We describe an implementation of the Deformable Parts Model [1] that operates in a user-defined time-frame. Our implementation uses a variety of mechanism to trade-off speed against accuracy. Our implementation can detect all 20 PASCAL 2007 objects simultaneously at 30Hz with an mAP of 0.26. At 15Hz, its mAP is 0.30; and at 100Hz, its mAP is 0.16. By comparison the reference implementation of [1] runs at 0.07Hz and mAP of 0.33 and a fast GPU implementation runs at 1Hz. Our technique is over an order of magnitude faster than the previous fastest DPM implementation. Our implementation exploits a series of important speedup mechanisms. We use the cascade framework of [3] and the vector quantization technique of [2]. To speed up feature computation, we compute HOG features at few scales, and apply many interpolated templates. A hierarchical vector quantization method is used to compress HOG features for fast template evaluation. An object proposal step uses hash-table methods to identify locations where evaluating templates would be most useful; these locations are inserted into a priority queue, and processed in a detection phase. Both proposal and detection phases have an any-time property. Our method applies to legacy templates, and no retraining is required.

AB - We describe an implementation of the Deformable Parts Model [1] that operates in a user-defined time-frame. Our implementation uses a variety of mechanism to trade-off speed against accuracy. Our implementation can detect all 20 PASCAL 2007 objects simultaneously at 30Hz with an mAP of 0.26. At 15Hz, its mAP is 0.30; and at 100Hz, its mAP is 0.16. By comparison the reference implementation of [1] runs at 0.07Hz and mAP of 0.33 and a fast GPU implementation runs at 1Hz. Our technique is over an order of magnitude faster than the previous fastest DPM implementation. Our implementation exploits a series of important speedup mechanisms. We use the cascade framework of [3] and the vector quantization technique of [2]. To speed up feature computation, we compute HOG features at few scales, and apply many interpolated templates. A hierarchical vector quantization method is used to compress HOG features for fast template evaluation. An object proposal step uses hash-table methods to identify locations where evaluating templates would be most useful; these locations are inserted into a priority queue, and processed in a detection phase. Both proposal and detection phases have an any-time property. Our method applies to legacy templates, and no retraining is required.

KW - Fast Deformable Parts Model

KW - Fast Object Detection

KW - Real-time Object Detection

UR - http://www.scopus.com/inward/record.url?scp=84906514441&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84906514441&partnerID=8YFLogxK

U2 - 10.1007/978-3-319-10590-1_5

DO - 10.1007/978-3-319-10590-1_5

M3 - Conference contribution

AN - SCOPUS:84906514441

SN - 9783319105895

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 65

EP - 79

BT - Computer Vision, ECCV 2014 - 13th European Conference, Proceedings

PB - Springer-Verlag

ER -