The Fastest Deformable Part Model for Object Detection

By Junjie Yan, Zhen Lei, Longyin Wen, S. Li. 2014

289 citations recorded by the source corpus.

This paper solves the speed bottleneck of deformable part model (DPM), while maintaining the accuracy in detection on challenging datasets. Three prohibitive steps in cascade version of DPM are accelerated, including 2D correlation between root filter and feature map, cascade part pruning and HOG feature extraction. For 2D correlation, the root filter is constrained to be low rank, so that 2D correlation can be calculated by more efficient linear combination of 1D correlations. A proximal gradient algorithm is adopted to progressively learn the low rank filter in a discriminative manner. For cascade part pruning, neighborhood aware cascade is proposed to capture the dependence in neighborhood regions for aggressive pruning. Instead of explicit computation of part scores, hypotheses can be pruned by scores of neighborhoods under the first order approximation. For HOG feature extraction, look-up tables are constructed to replace expensive calculations of orientation partition and magnitude with simpler matrix index operations. Extensive experiments show that (a) the proposed method is 4 times faster than the current fastest DPM method with similar accuracy on Pascal VOC, (b) the proposed method achieves state-of-the-art accuracy on pedestrian and face detection task with frame-rate speed.

Subfields: Computer Vision and Pattern Recognition

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

Citations in the corpus

You Only Look Once: Unified, Real-Time Object Detection (2015, background): “[5] [37] [30] [14] [17] [27] However, only Sadeghi et al.”
From Facial Parts Responses to Face Detection: A Deep Learning Approach (2015, methodology): “We compare Faceness-Net against all published methods [37, 23, 3, 35, 18, 20, 17, 28, 40, 13] in the FDDB.”
Scale-Aware Face Detection (2017, methodology): “These methods achieves better performance than Viola-Jones based methods on standard benchmarks such as AFW [42] and FDDB [16], and progressively become more efﬁcient, including [42, 25, 35, 10].”
Factors in Finetuning Deep Model for Object Detection with Long-Tail Distribution (2016, methodology): “When using the hand-crafted features, the same feature extraction mechanism is used for all object classes [2, 41, 10].”
Adaptive Feeding: Achieving Fast and Accurate Detections by Adaptively Combining Object Detectors (2017, methodology): “After DPM was invented, many DPM-based detectors [24, 30, 6] focused on optimizing different parts in the pipeline.”
Multi-task Generative Adversarial Network for Detecting Small Objects in the Wild (2020, unknown): “Existing face detectors can also be broadly divided into two categories: handcrafted feature based methods (Viola and Jones 2004; Yan et al. 2014, 2013; Zhang et al. 2017c) and CNN-based methods (Cai et al. 2016; Jiang and Learned-Miller 2017; Wan et al. 2016; Zhu et al. 2017a; Bai and Ghanem 2017;…”
S^3FD: Single Shot Scale-Invariant Face Detector (2017, methodology): “Besides the cascade structure, [30, 51, 63] introduce deformable part models (DPM) into face detection tasks and achieve remarkable performance.”
Finding Tiny Faces in the Wild with Generative Adversarial Network (2018, methodology): “Existing face detection methods can be broadly categorized as handcrafted feature based methods [24, 29, 30] and CNN-based methods [34, 2, 14, 25, 37, 1].”
PyramidBox: A Context-assisted Single Shot Face Detector (2018, methodology): “Then, [8,9,10] introduce deformable part models (DPM) into face detection tasks by modeling the relationship of deformable facial parts.”
Leaving Some Stones Unturned: Dynamic Feature Prioritization for Activity Detection in Streaming Video (2016, methodology): “We use the DPM detector [49], which has the advantage of near real-time detection [38] using a single thread, whereas R-CNN relies heavily on parallel computation and hardware acceleration [50].”
Learning coarse-to-fine sparselets for efficient object detection and scene classification (2015, background): “In recent years, since the seminal work of [4, 5] and with the success of deformable part models (DPMs) [6, 7], part model-based methods have demonstrated impressive performance for object detection [6-13] and scene classification [14-21].”
Single-Pedestrian Detection Aided by Multi-pedestrian Detection (2013, background): “Fast object detection algorithms often gain high interests [4], [12], [19], [40], [90].”
Ten Years of Pedestrian Detection, What Have We Learned? (2014, background): “For pedestrian detection the results are competitive, but not salient ( LatSvm [50,12], MultiResC [33], MT-DPM [39]).”
Quantization Mimic: Towards Very Tiny CNN for Object Detection (2018, background): “The target of object detection [5,6,7,8,9,10] is to locate and classify the objects in images.”
Face Detection without Bells and Whistles (2014, result): “[32] independently reported results DPM for the AFW dataset which are consistent to our results.”
Joint Cascade Face Detection and Alignment (2014, result): “On FDDB dataset, we compare with all the published results in the platform [26,32,24,14,23,12,16,8,25,18,30,13,15].”
30Hz Object Detection with DPM V5 (2014, background): “[23] is not included in the table as its running time (0.”