Fast Feature Pyramids for Object Detection

Piotr DollárRon AppelSerge J. BelongieP. Perona

article2014TPAMI2,020 citations

Demonstrates how to construct fast multi-scale feature pyramids by extrapolating intermediate levels from octave-spaced scales using natural image statistics, enabling real-time object and pedestrian detection with minimal accuracy loss.

Listen

Visual object detection systems, such as those used in autonomous driving, mobile devices, and robotics, require high accuracy while operating under strict real-time and low-power computational constraints. Over recent years, detection algorithms achieved dramatic improvements in accuracy by extracting rich visual representations across finely sampled scale pyramids, but this drastically increased processing time from real-time rates to multiple seconds per image. The article sets out to demonstrate that densely sampled visual feature pyramids can be rapidly estimated via mathematical extrapolation from coarsely sampled scales, eliminating the conventional trade-off between computational speed and detection accuracy.

The authors conducted a comprehensive statistical and empirical analysis grounded in the fractal properties of natural images. They evaluated a broad class of low-level visual features across thousands of natural and pedestrian images to model feature behavior across scale changes. They integrated this scaling model into three distinct detection systems: Aggregated Channel Features, Integral Channel Features, and Deformable Part Models. These systems were evaluated on standard benchmark datasets, including INRIA, Caltech, TUD-Brussels, and ETH for pedestrian detection, and PASCAL VOC for general multi-class object detection.

The article established several critical findings. First, low-level visual features scale predictably across image resolutions according to a consistent power law, allowing feature channels computed at sparse intervals of one octave to accurately predict intermediate scales. Second, feature computation costs are reduced by roughly an order of magnitude, requiring only 33% more computation than single-scale extraction rather than the dense multi-scale baseline. Third, the new Aggregated Channel Features detector achieved real-time speeds exceeding 30 frames per second on standard resolution imagery on a single computer processor, compared to 12 frames per second for exact multi-scale extraction. Fourth, this massive speedup caused negligible loss in accuracy, yielding an average miss rate across pedestrian datasets of 41% with fast feature pyramids compared to 40% with exact pyramids, while Deformable Part Models on general object detection saw only a 2% drop in mean average precision.

These findings show that computing redundant, high-resolution features across every scale is unnecessary. For technical leaders and engineering teams, this method enables real-time, high-accuracy computer vision on cost-effective, standard hardware without requiring expensive graphics processing units. Practitioners should adopt fast feature pyramid architectures in sliding-window visual pipelines, calibrate scaling exponents on representative domain data, and combine these pyramids with optimized classification cascades for maximum frame rate gains. Decision-makers should note that the underlying power-law assumptions apply specifically to natural scenes with broad visual spectra; caution is warranted when applying the method to synthetic environments, extreme magnifications, or narrow-band periodic textures where the approximation breaks down.

California Institute of Technology Cornell University Microsoft detection algorithm detection rates image features fast feature pyramids object detection pedestrian detection

pdollar/toolbox Dollár et al (2014).pdf

Paper: Object Detection with Discriminatively Trained Part-Based Models, Pedro F. Felzenszwalb et al. (2010). The source incorporates Deformable Part Models, so their part-based detector and multi-scale framework clarify one of the systems it accelerates.

Paper: Feature Pyramid Networks for Object Detection, Tsung-Yi Lin et al. (2017). Feature Pyramid Networks carry multi-scale detection forward by building semantic feature pyramids inside a CNN rather than extrapolating features between image scales.
Paper: NAS-FPN: Learning Scalable Feature Pyramid Architecture for Object Detection, Golnaz Ghiasi et al. (2019). NAS-FPN extends feature-pyramid design by using architecture search to discover and scale cross-level connections beyond manually specified pyramid structures.
Paper: EfficientDet: Scalable and Efficient Object Detection, Mingxing Tan et al. (2020). EfficientDet continues the feature-pyramid line with a learned bidirectional fusion network and compound scaling for detectors across compute budgets.

Cover for Fast Feature Pyramids for Object Detection

Abstract

Multi-resolution image features may be approximated via extrapolation from nearby scales, rather than being computed explicitly. This fundamental insight allows us to design object detection algorithms that are as accurate, and considerably faster, than the state-of-the-art. The computational bottleneck of many modern detectors is the computation of features at every scale of a finely-sampled image pyramid. Our key insight is that one may compute finely sampled feature pyramids at a fraction of the cost, without sacrificing performance: for a broad family of features we find that features computed at octave-spaced scale intervals are sufficient to approximate features on a finely-sampled pyramid. Extrapolation is inexpensive as compared to direct feature computation. As a result, our approximation yields considerable speedups with negligible loss in detection accuracy. We modify three diverse visual recognition systems to use fast feature pyramids and show results on both pedestrian detection (measured on the Caltech, INRIA, TUD-Brussels and ETH datasets) and general object detection (measured on the PASCAL VOC). The approach is general and is widely applicable to vision algorithms requiring fine-grained multi-scale analysis. Our approximation is valid for images with broad spectra (most natural images) and fails for images with narrow band-pass spectra (e.g. periodic textures).

Topics

detection algorithm detection rates image features object detection algorithms pedestrian detection vision algorithms multiresolution images laplacian pyramid

Keywords

fast feature pyramids object detection pedestrian detection multiscale detection approximate multiscale features integral channel features octave-spaced feature pyramid HOG pyramids gradient histograms

1 Introduction
2 Related Work
3 Multiscale Gradient Histograms
3.1 Gradient Histograms in Upsampled Images
3.2 Gradient Histograms in Downsampled Images
3.3 Histograms of Normalized Gradients
4 Statistics of Multiscale Features
4.1 Power Law Governs Feature Scaling
4.2 Estimating λ
4.3 Deviation for Individual Images
4.4 Miscellanea
5 Fast Feature Pyramids
5.1 Feature Channel Scaling
5.2 Fast Feature Pyramids
5.3 Complexity Analysis
6 Applications to Object Detection
6.1 Aggregated Channel Features (ACF)
6.2 Integral Channel Features (ICF)
6.3 Deformable Part Models (DPM)
7 Conclusion
Acknowledgments
References

Knowls

Knowl 1 — Power Law Governing Feature Channel Scaling
theoretical result
Let $\Omega$ be any shift-invariant feature mapping that transforms an image $I$ into a multi-channel feature map $C = \Omega(I)$ with spatial dimensions $h \times w$ and $k$ layers. Let $I_s$ denote the image at scale $s$ (with dimensions $s h \times s w$ ), and let $C_s = \Omega(I_s)$ denote the channel computed at scale $s$ . The global mean feature response across all spatial locations and layers is defined as:

$f_\Omega(I_s) = \frac{1}{s^2 h w k} \sum_{i=1}^{s h} \sum_{j=1}^{s w} \sum_{l=1}^{k} C_s(i, j, l)$

Because natural images exhibit scale-invariant, fractal power spectra, the ratio of mean feature responses computed at two arbitrary scales $s_1$ and $s_2$ follows a power law:

$\frac{f_\Omega(I_{s_1})}{f_\Omega(I_{s_2})} = \left(\frac{s_1}{s_2}\right)^{-\lambda_\Omega} + \mathcal{E}$

where $\lambda_\Omega$ is a channel-specific scaling exponent and $\mathcal{E}$ is a zero-mean random deviation ( $\mathbb{E}[\mathcal{E}] \approx 0$ ) whose variance increases gradually with the scale ratio $|\log_2(s_1/s_2)|$ . The same scaling relationship holds locally for sub-windows and local histograms.
Knowl 2 — Feature Channel Approximation via Power-Law Rescaling
equation
Given a precomputed feature channel map $C_{s'} = \Omega(I_{s'})$ at scale $s'$ , the feature channel map $C_s = \Omega(I_s)$ at a nearby scale $s$ is approximated directly without evaluating $\Omega(I_s)$ by:

$C_s \approx R\left(C_{s'}, \frac{s}{s'}\right) \cdot \left(\frac{s}{s'}\right)^{-\lambda_\Omega}$

where $R(C, \alpha)$ denotes the spatial resampling (e.g., bilinear interpolation) of channel map $C$ by a scale factor of $\alpha$ , and $\lambda_\Omega$ is the characteristic power-law scaling exponent for channel type $\Omega$ . The approximation error is minimized when $C$ is spatially aggregated (downsampled or smoothed) over local pixel blocks.
Knowl 3 — Fast Feature Pyramid Construction
algorithm
The fast feature pyramid algorithm computes exact feature maps only at sparsely sampled octave intervals and approximates all intermediate scale levels via spatial resampling and power-law scaling.
Input: Input image $I$ , scale set $S = \{s_1, s_2, \dots, s_M\}$ sampled logarithmically with $m$ scales per octave, feature extraction operator $\Omega$ , scaling exponent $\lambda_\Omega$
Output: Multiscale feature pyramid $\{C_s\}_{s \in S}$
OctaveScales $\leftarrow \{s \in S \mid \log_2(s) \in \mathbb{Z}\}$
for each $s' \in \text{OctaveScales}$ do
     $I_{s'} \leftarrow R(I, s')$
     $C_{s'} \leftarrow \Omega(I_{s'})$
end for
for each $s \in S$ do
    if $s \in \text{OctaveScales}$ then
        continue
    end if
     $s^* \leftarrow \arg\min_{u \in \text{OctaveScales}} |\log_2(s) - \log_2(u)|$
     $\alpha \leftarrow s / s^*$
     $C_s \leftarrow R(C_{s^*}, \alpha) \cdot \alpha^{-\lambda_\Omega}$
end for
return $\{C_s\}_{s \in S}$
Knowl 4 — Computational Complexity of Exact vs. Fast Feature Pyramids
theoretical result
Assuming the feature extractor $\Omega$ has a cost linear in the number of pixels of an $n \times n$ image, constructing an exact feature pyramid with $m$ scales per octave over an infinite geometric progression of scales has a computational cost of:

$\sum_{k=0}^{\infty} n^2 2^{-2k/m} = \frac{n^2}{1 - 4^{-1/m}} \approx \frac{m n^2}{\ln 4}$

For typical sampling densities of $m = 8$ to $12$ scales per octave, exact computation requires approximately $6 n^2$ to $9 n^2$ operations.

In contrast, the fast feature pyramid evaluates $\Omega$ only once per octave ( $m = 1$ ), which requires:

$\sum_{k=0}^{\infty} n^2 4^{-k} = \frac{4}{3} n^2$

This represents only a $33\%$ overhead over evaluating features at the single base scale $n^2$ . Approximating the intermediate $m-1$ scales via 2D spatial resampling $R$ introduces minimal overhead, particularly when channels are spatially aggregated.
Knowl 5 — Aggregated Channel Features (ACF) Detector
model/method
The Aggregated Channel Features (ACF) object detector operates on 10 channel maps computed from an input image $I$ :
1. 1 normalized gradient magnitude channel: $\widetilde{M}(i, j) = M(i, j) / (\bar{M}(i, j) + 0.005)$ , where $M$ is gradient magnitude and $\bar{M}$ is obtained by convolving $M$ with an $L_1$ -normalized $11 \times 11$ triangle filter.
2. 6 orientation-quantized gradient histogram channels.
3. 3 LUV color channels.
Prior to channel computation, $I$ is smoothed with a $[1\ 2\ 1]/4$ filter. The 10 channels are divided into non-overlapping $4 \times 4$ spatial blocks, summed within each block, and post-smoothed with a $[1\ 2\ 1]/4$ filter.

Features are single-pixel lookups into the aggregated channels. For a $128 \times 64$ detection window, this yields $(128/4) \times (64/4) \times 10 = 5120$ candidate features. A boosted classifier combining 2048 depth-two decision trees is trained using AdaBoost with multiple bootstrapping rounds. During sliding-window evaluation, the detector steps by 4 pixels with 8 scales per octave, running at over 30 frames per second on $640 \times 480$ images on a single CPU.
Knowl 6 — Empirical Scaling Exponents Across Standard Visual Features
empirical result
Least-squares estimation of $\log_2(\mu_s) = a'_\Omega - \lambda_\Omega \log_2(s)$ over 24 downsampled scales across 3 octaves yields distinct scaling exponents $\lambda_\Omega$ and mean absolute fitting errors $|\mathbb{E}[\mathcal{E}]|$ on natural and pedestrian image ensembles:
- Grayscale pixel intensities: $\lambda = 0.000$ , $|\mathbb{E}[\mathcal{E}]| = 0.000$
- HOG ( $4 \times 4$ spatial bins, 36 channels): $\lambda = 0.078$ , $|\mathbb{E}[\mathcal{E}]| = 0.004$ (natural), $0.019$ (pedestrian)
- Normalized gradient histograms (6 bins): $\lambda = 0.101$ , $|\mathbb{E}[\mathcal{E}]| = 0.006$ (natural), $0.026$ (pedestrian)
- Local standard deviation ( $5 \times 5$ window): $\lambda = 0.358$ , $|\mathbb{E}[\mathcal{E}]| = 0.025$ (natural), $0.044$ (pedestrian)
- Unnormalized gradient histograms (6 bins): $\lambda = 0.406$ , $|\mathbb{E}[\mathcal{E}]| = 0.018$ (natural), $0.037$ (pedestrian)
- Difference of Gaussians (DoG, $\sigma_{in} = 0.71, \sigma_{out} = 1.14$ ): $\lambda = 0.974$ , $|\mathbb{E}[\mathcal{E}]| = 0.013$ (natural), $0.019$ (pedestrian)
For all tested channels, the standard deviation of single-image deviation at scale $s = 1/2$ satisfies $\sigma_{1/2} < 0.20$ .

Knowl 7 — Pedestrian Detection Benchmark Comparison of Exact vs. Fast Pyramids

data/table

Log-average miss rates (MR, in percent, lower is better) evaluated between $10^{-2}$ and $10^0$ false positives per image across four standard pedestrian datasets:

Method	INRIA	Caltech	TUD-Brussels	ETH	Mean
Shapelet	82	91	95	91	90
Viola-Jones (VJ)	72	95	95	90	88
PoseInv	80	86	88	92	87
HikSvm	43	73	83	72	68
HOG	46	68	78	64	64
HogLbp	39	68	82	55	61
MF	36	68	73	60	59
PLS	40	62	71	55	57
MF+CSS	25	61	60	61	52
LatSvmV2	20	63	70	51	51
FPDW	21	57	63	60	50
ChnFtrs	22	56	60	57	49
Crosstalk	19	54	58	52	46
ICF-Exact	18	48	53	50	42
ICF (approximate)	19	51	55	56	45
ACF-Exact	17	43	50	50	40
ACF (approximate)	17	45	52	51	41

Approximating 7 of every 8 pyramid scales incurs only a $1\%$ increase in mean MR for ACF ( $40\%$ to $41\%$ ) and a $3\%$ increase for ICF ( $42\%$ to $45\%$ ), while accelerating ACF to over 30 fps on $640 \times 480$ images.

Knowl 8 — Performance of Deformable Part Models Using Fast Feature Pyramids on PASCAL VOC 2007

data/table

Average Precision (AP, in percent, higher is better) for Deformable Part Models on the 20 object classes of PASCAL VOC 2007 comparing exact HOG pyramids (DPM) to pyramids where 9 out of 10 scales per octave are approximated via the power law ( $\sim$ DPM):

Method	plane	bike	bird	boat	bottle	bus	car	cat	chair	cow
DPM	26.3	59.4	2.3	10.2	21.2	46.2	52.2	7.9	15.9	17.4
$\sim$ DPM	24.1	54.7	1.6	9.8	20.0	42.1	50.1	8.0	13.8	16.7
Method	table	dog	horse	moto	person	plant	sheep	sofa	train	tv
DPM	10.9	2.9	53.4	37.6	38.2	4.9	16.6	29.7	38.2	40.8
$\sim$ DPM	8.9	2.5	49.4	38.3	36.0	4.2	14.9	24.4	35.8	35.0

Mean AP across all 20 classes is $26.6\%$ with exact pyramids and $24.5\%$ with fast feature pyramids, demonstrating that the power-law approximation generalizes to multi-component deformable models with a minor ( $2.1\%$ ) drop in detection performance.

Knowl 9 — Detector Rescaling via Feature Power Law in Integral Channel Features
model/method
In Integral Channel Features (ICF), feature values are defined as rectangular sums over channel maps: $f(I) = A f_\Omega(I)$ , where $A$ is the rectangle area and $f_\Omega(I)$ is the average channel value. Rather than rescaling channel images, detector responses at scale offset $\alpha = s/s'$ can be approximated directly from integral channels computed at reference scale $s'$ by rescaling the classifier itself:
1. Each rectangular feature bounding box of width $w$ and height $h$ is rescaled to dimensions $\alpha w \times \alpha h$ .
2. The rectangle sum evaluated on the reference integral image is scaled by $\alpha^{-\lambda_\Omega}$ .
A hybrid implementation computes exact integral feature channels once per octave and scales the detector model to evaluate detection windows within half an octave of each reference level.
Knowl 10 — Scope and Failure Modes of Fast Feature Pyramid Approximation
limitation
The power-law feature approximation relies on the broad fractal spatial spectra of natural scenes and fails or degrades in three specific regimes:
1. Narrow band-pass spectra: Highly structured, periodic textures (such as certain Brodatz textures) with high-frequency energy do not follow the scale-invariance power law, causing significant underestimation of feature energy upon downsampling.
2. Highly sparse channels: Channels where the feature value is frequently near zero ( $f_\Omega(I) \approx 0$ , e.g., the output of a sparse classifier) exhibit high variance in the scaling ratio and cannot be reliably extrapolated.
3. Extreme scale regimes: Scales far outside natural optical viewing ranges (e.g., severe microscopic magnification) break the scale stationarity of image statistics.

Coverage note — None was omitted; all contributed theory, equations, algorithms, detector models (ACF, ICF, DPM), empirical scaling parameter fits, and full benchmark evaluation tables from the paper are represented.

References

1.D. Hubel and T. Wiesel, “Receptive fields and functional architecture of monkey striate cortex,” Journal of Physiology, 1968.
2.C. Malsburg, “Self-organization of orientation sensitive cells in the striate cortex,” Biological Cybernetics, vol. 14, no. 2, 1973.
3.L. Maffei and A. Fiorentini, “The visual cortex as a spatial frequency analyser,” Vision Research, vol. 13, no. 7, 1973.
4.P. Burt and E. Adelson, “The laplacian pyramid as a compact image code,” IEEE Transactions on Communications, 1983.
5.J. Daugman, “Uncertainty relation for resolution in space, spatial frequency, and orientation optimized by two-dimensional visual cortical filters,” Journal of the Optical Society of America A, 1985.
6.J. Koenderink and A. Van Doorn, “Representation of local geometry in the visual system,” Biological cybernetics, vol. 55, no. 6, 1987.
7.D. J. Field, “Relations between the statistics of natural images and the response properties of cortical cells,” Journal of the Optical Society of America A, vol. 4, pp. 2379–2394, 1987.
8.S. Mallat, “‘A theory for multiresolution signal decomposition: the wavelet representation,” PAMI, vol. 11, no. 7, 1989.
9.P. Vaidyanathan, “‘Multirate digital filters, filter banks, polyphase networks, and applications: A tutorial,” Proceedings of the IEEE, vol. 78, no. 1, 1990.
10.M. Vetterli, “‘A theory of multirate filter banks,” IEEE Conference on Acoustics, Speech and Signal Processing, vol. 35, no. 3, 1987.
11.E. Simoncelli and E. Adelson, “Noise removal via bayesian wavelet coring,” in ICIP, vol. 1, 1996.
12.W. T. Freeman and E. H. Adelson, “The design and use of steerable filters,” PAMI, vol. 13, pp. 891–906, 1991.
13.J. Malik and P. Perona, “‘Preattentive texture discrimination with early vision mechanisms,” Journal of the Optical Society of America A, vol. 7, pp. 923–932, May 1990.
14.D. Jones and J. Malik, “Computational framework for determining stereo correspondence from a set of linear spatial filters,” Image and Vision Computing, vol. 10, no. 10, pp. 699–708, 1992.
15.E. Adelson and J. Bergen, “Spatiotemporal energy models for the perception of motion,” Journal of the Optical Society of America A, vol. 2, no. 2, pp. 284–299, 1985.
16.Y. Weiss and E. Adelson, “‘A unified mixture framework for motion segmentation: Incorporating spatial coherence and estimating the number of models,” in CVPR, 1996.
17.L. Itti, C. Koch, and E. Niebur, “‘A model of saliency-based visual attention for rapid scene analysis,” PAMI, vol. 20, no. 11, 1998.
18.P. Perona and J. Malik, “Detecting and localizing edges composed of steps, peaks and roofs,” in ICCV, 1990.
19.M. Lades, J. Vorbruggen, J. Buhmann, J. Lange, C. von der Malsburg, R. Wurtz, and W. Konen, “‘Distortion invariant object recognition in the dynamic link architecture,” IEEE Transactions on Computers, vol. 42, no. 3, pp. 300–311, 1993.
20.D. G. Lowe, “‘Object recognition from local scale-invariant features,” in ICCV, 1999.
21.N. Dalal and B. Triggs, “‘Histograms of oriented gradients for human detection,” in CVPR, 2005.
22.R. De Valois, D. Albrecht, and L. Thorell, “‘Spatial frequency selectivity of cells in macaque visual cortex,” Vision Research, vol. 22, no. 5, pp. 545–559, 1982.
23.Y. LeCun, P. Haffner, L. Bottou, and Y. Bengio, “‘Gradient-based learning applied to document recognition,” in Proc. of IEEE, 1998.
24.M. Riesenhuber and T. Poggio, “‘Hierarchical models of object recognition in cortex,” Nature Neuroscience, vol. 2, 1999.
25.D. G. Lowe, “‘Distinctive image features from scale-invariant keypoints,” IJCV, vol. 60, no. 2, pp. 91–110, 2004.
26.A. Krizhevsky, I. Sutskever, and G. Hinton, “Imagenet classification with deep convolutional neural networks,” in NIPS, 2012.
27.P. Viola and M. Jones, “‘Rapid object detection using a boosted cascade of simple features,” in CVPR, 2001.
28.P. Viola, M. Jones, and D. Snow, “Detecting pedestrians using patterns of motion and appearance,” IJCV, vol. 63(2), 2005.
29.P. Doll'ar, Z. Tu, P. Perona, and S. Belongie, “Integral channel features,” in BMVC, 2009.
30.R. Benenson, M. Mathias, R. Timofte, and L. Van Gool, “‘Pedestrian detection at 100 frames per second,” in CVPR, 2012.
31.P. Doll'ar, C. Wojek, B. Schiele, and P. Perona, “‘Pedestrian detection: An evaluation of the state of the art,” PAMI, vol. 99, 2011.
32.www.vision.caltech.edu/Image_Datasets/CaltechPedestrians/.
33.D. L. Ruderman and W. Bialek, “Statistics of natural images: Scaling in the woods,” Physical Review Letters, vol. 73, no. 6, pp. 814–817, Aug 1994.
34.E. Switkes, M. Mayer, and J. Sloan, “‘Spatial frequency analysis of the visual environment: anisotropy and the carpentered environment hypothesis,” Vision Research, vol. 18, no. 10, 1978.
35.P. Felzenszwalb, R. Girshick, D. McAllester, and D. Ramanan, “‘Object detection with discriminatively trained part based models,” PAMI, vol. 32, no. 9, pp. 1627–1645, 2010.
36.C. Wojek, S. Walk, and B. Schiele, “‘Multi-cue onboard pedestrian detection,” in CVPR, 2009.
37.A. Ess, B. Leibe, and L. Van Gool, “‘Depth and appearance for mobile scene analysis,” in ICCV, 2007.
38.M. Everingham, L. Van Gool, C. K. I. Williams, J. Winn, and A. Zisserman, “‘The PASCAL visual object classes (VOC) challenge,” IJCV, vol. 88, no. 2, pp. 303–338, Jun. 2010.
39.P. Doll'ar, S. Belongie, and P. Perona, “‘The fastest pedestrian detector in the west,” in BMVC, 2010.
40.P. Doll'ar, R. Appel, and W. Kienzle, “‘Crosstalk cascades for frame-rate pedestrian detection,” in ECCV, 2012.
41.T. Lindeberg, “Scale-space for discrete signals,” PAMI, vol. 12, no. 3, pp. 234–254, 1990.
42.J. L. Crowley, O. Riff, and J. H. Piater, “‘Fast computation of characteristic scale using a half-octave pyramid,” in International Conference on Scale-Space Theories in Computer Vision, 2002.
43.R. S. Eaton, M. R. Stevens, J. C. McBride, G. T. Foil, and M. S. Snorrason, “‘A systems view of scale space,” in ICVS, 2006.
44.P. Felzenszwalb, R. Girshick, and D. McAllester, “‘Cascade object detection with deformable part models,” in CVPR, 2010.
45.M. Pedersoli, A. Vedaldi, and J. Gonzalez, “‘A coarse-to-fine approach for fast deformable object detection,” in CVPR, 2011.
46.C. H. Lampert, M. B. Blaschko, and T. Hofmann, “‘Efficient subwindow search: A branch and bound framework for object localization,” PAMI, vol. 31, pp. 2129–2142, Dec 2009.
47.L. Bourdev and J. Brandt, “‘Robust object detection via soft cascade,” in CVPR, 2005.
48.C. Zhang and P. Viola, “‘Multiple-instance pruning for learning efficient cascade detectors,” in NIPS, 2007.
49.J. Šochman and J. Matas, “‘Waldboost - learning for time constrained sequential detection,” in CVPR, 2005.
50.H. Masnadi-Shirazi and N. Vasconcelos, “‘High detection-rate cascades for real-time object detection,” in ICCV, 2007.
51.F. Fleuret and D. Geman, “‘Coarse-to-fine face detection,” IJCV, vol. 41, no. 1-2, pp. 85–107, 2001.
52.P. Felzenszwalb and D. Huttenlocher, “‘Efficient matching of pictorial structures,” in CVPR, 2000.
53.C. Papageorgiou and T. Poggio, “‘A trainable system for object detection,” IJCV, vol. 38, no. 1, pp. 15–33, 2000.
54.M. Weber, M. Welling, and P. Perona, “‘Unsupervised learning of models for recognition,” in ECCV, 2000.
55.S. Agarwal and D. Roth, “‘Learning a sparse representation for object detection,” in ECCV, 2002.
56.R. Fergus, P. Perona, and A. Zisserman, “‘Object class recognition by unsupervised scale-invariant learning,” in CVPR, 2003.
57.B. Leibe, A. Leonardis, and B. Schiele, “‘Robust object detection with interleaved categorization and segmentation,” IJCV, vol. 77, no. 1-3, pp. 259–289, May 2008.
58.C. Gu, J. J. Lim, P. Arbelaez, and J. Malik, “‘Recognition using regions,” in CVPR, 2009.
59.B. Alexe, T. Deselaers, and V. Ferrari, “What is an object?” in CVPR, 2010.
60.C. Wojek, G. Dorko, A. Schulz, and B. Schiele, “‘Sliding-windows for rapid object class localization: A parallel technique,” in DAGM, 2008.
61.L. Zhang and R. Nevatia, “‘Efficient scan-window based object detection using gpgpu,” in Visual Computer Vision on GPU’s (CVGPU), 2008.
62.B. Bilgic, “‘Fast human detection with cascaded ensembles,” Master’s thesis, MIT, February 2010.
63.Q. Zhu, S. Avidan, M. Yeh, and K. Cheng, “‘Fast human detection using a cascade of histograms of oriented gradients,” in CVPR, 2006.
64.F. M. Porikli, “Integral histogram: A fast way to extract histograms in cartesian spaces,” in CVPR, 2005.
65.A. Ess, B. Leibe, K. Schindler, and L. Van Gool, “‘Robust multi-person tracking from a mobile platform,” PAMI, vol. 31, pp. 1831–1846, 2009.
66.M. Bajracharya, B. Moghaddam, A. Howard, S. Brennan, and L. H. Matthies, “‘A fast stereo-based system for detecting and tracking pedestrians from a moving vehicle,” The International Journal of Robotics Research, vol. 28, 2009.
67.D. L. Ruderman, “The statistics of natural images,” Network: Computation in Neural Systems, vol. 5, no. 4, pp. 517–548, 1994.
68.S. G. Ghurye, “‘A characterization of the exponential function,” The American Mathematical Monthly, vol. 64, no. 4, 1957.
69.J. Friedman, T. Hastie, and R. Tibshirani, “‘Additive logistic regression: a statistical view of boosting,” The Annals of Statistics, vol. 38, no. 2, pp. 337–374, 2000.
70.P. Sabzmeydani and G. Mori, “‘Detecting pedestrians by learning shapelet features,” in CVPR, 2007.
71.Z. Lin and L. S. Davis, “‘A pose-invariant descriptor for human detection and segmentation,” in ECCV, 2008.
72.S. Maji, A. Berg, and J. Malik, “‘Classification using intersection kernel SVMs is efficient,” in CVPR, 2008.
73.X. Wang, T. X. Han, and S. Yan, “‘An hog-lbp human detector with partial occlusion handling,” in ICCV, 2009.
74.C. Wojek and B. Schiele, “‘A performance evaluation of single and multi-feature people detection,” in DAGM, 2008.
75.W. Schwartz, A. Kembhavi, D. Harwood, and L. Davis, “‘Human detection using partial least squares analysis,” in ICCV, 2009.
76.S. Walk, N. Majer, K. Schindler, and B. Schiele, “‘New features and insights for pedestrian detection,” in CVPR, 2010.
77.D. Park, D. Ramanan, and C. Fowlkes, “‘Multiresolution models for object detection,” in ECCV, 2010.

Citation

MLA

Dollar, P., et al. “Fast Feature Pyramids for Object Detection”. IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 36, no. 8, 2014, pp. 1532–45, https://doi.org/10.1109/TPAMI.2014.2300479.

APA

Dollar, P., Appel, R., Belongie, S., & Perona, P. (2014). Fast Feature Pyramids for Object Detection. IEEE Transactions on Pattern Analysis and Machine Intelligence, 36(8), 1532–1545. https://doi.org/10.1109/TPAMI.2014.2300479

Chicago

Dollar, P., R. Appel, S. Belongie, and P. Perona. 2014. “Fast Feature Pyramids for Object Detection”. IEEE Transactions on Pattern Analysis and Machine Intelligence 36 (8): 1532–45. https://doi.org/10.1109/TPAMI.2014.2300479.

Harvard

Dollar, P. et al. (2014) “Fast Feature Pyramids for Object Detection”, IEEE Transactions on Pattern Analysis and Machine Intelligence, 36(8), pp. 1532–1545. Available at: https://doi.org/10.1109/TPAMI.2014.2300479.

Vancouver

1. Dollar P, Appel R, Belongie S, Perona P (2014) Fast Feature Pyramids for Object Detection. IEEE Transactions on Pattern Analysis and Machine Intelligence 36:1532–1545

BibTeX

@article{Dollar_2014, title={Fast Feature Pyramids for Object Detection}, volume={36}, ISSN={2160-9292}, url={http://dx.doi.org/10.1109/TPAMI.2014.2300479}, DOI={10.1109/tpami.2014.2300479}, number={8}, journal={IEEE Transactions on Pattern Analysis and Machine Intelligence}, publisher={Institute of Electrical and Electronics Engineers (IEEE)}, author={Dollar, Piotr and Appel, Ron and Belongie, Serge and Perona, Pietro}, year={2014}, month=Aug, pages={1532–1545} }

Metadata:Crossref

Source Code

This paper has an official code repository available. Click below to access the source code.

View Repository

Access the Paper

This paper is available from its original source. Click below to access the PDF.

Open PDF

Fast Feature Pyramids for Object Detection

Abstract

Topics

Keywords

Table of Contents

Knowls

Knowl 1 — Power Law Governing Feature Channel Scaling

Knowl 2 — Feature Channel Approximation via Power-Law Rescaling

Knowl 3 — Fast Feature Pyramid Construction

Knowl 4 — Computational Complexity of Exact vs. Fast Feature Pyramids

Knowl 5 — Aggregated Channel Features (ACF) Detector

Knowl 6 — Empirical Scaling Exponents Across Standard Visual Features