Histograms of Oriented Gradients for Human Detection
2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'05), 2005 · paper · AI and machine learning
Studies feature sets for detecting people in images, using grids of histograms of oriented gradients with a linear classifier. These descriptors significantly outperform existing features, and the authors introduce a harder pedestrian dataset of over 1800 annotated images.
Checked: PDF from lear.inrialpes.fr, fetched 4 Oct 2026. Page numbers are those of that PDF.
1 finding
Wording Fig. 4 caption (e): "Reducing the 16 pixel margin around the 64×128 detection window decreases the performance by about 3%" versus §6.5: "Decreasing it from 16 to 8 pixels (48×112 detection window) decreases performance by 6% at 10−4 FPPW".
- p. 5 Figure 4 caption, item (e)
decreases the performance by about 3%
- p. 7 §6.5 Detector Window and Context
(48×112 detection window) decreases performance by 6%
- p. 7 Sec 6.5
decreases performance by 6%
- p. 7
decreases performance by 6% at 10−4 FPPW
Only the authors can say what is right here.
Judge's note: Both verified on the page 5 image and the page 7 crop. Fig. 4(e) plots three windows (64x128, 56x120, 48x112). The caption does not say which reduction or which FPPW its 3% refers to; it may mean the intermediate 56×120 window, but the paper does not say so. As printed, the same panel carries two different loss figures; the authors should say which case the caption describes.
Second check: another model agreed
Report a problem with this finding
10 minor (typos and the like; most checked by the judge only)
Typo §2 text reads "Gavrila & Philomen [8]" while reference [8] reads "D. M. Gavrila and V. Philomin"; the same author's name is spelled two ways.
- p. 1 §2 Previous Work
Gavrila & Philomen
- p. 8 Reference [8]
D. M. Gavrila and V. Philomin.
- p. 8 References, [8]
D. M. Gavrila and V. Philomin
- p. 8 Reference [8]
V. Philomin
Judge's note: Checked page 1 image (body text shows 'Philomen') and page 8 image (reference [8] shows 'Philomin'). Internal spelling inconsistency; the evident fix is to match the reference-list form.
Second check: none (judge only)
Report a problem with this finding
Typo Affiliation line reads "INRIA Rhône-Alps" (French accented 'Rhône' with English 'Alps'); probably intended 'Rhône-Alpes'.
- p. 1 Author affiliation line
INRIA Rhône-Alps
Only the authors can say what is right here.
Judge's note: Rests on outside knowledge: the institute/region is named 'Rhône-Alpes'. Page 1 image shows 'INRIA Rhône-Alps' as quoted. Nothing in the files states the official name (the email domain 'inrialpes.fr' does not settle it), so this is an author query.
Second check: none (judge only)
Report a problem with this finding
Typo Text has "Depoortere et al" where reference [2] has "V. de Poortere"; text has "Papageorgiou et al [18]" where reference [18] lists only two authors, "C. Papageorgiou and T. Poggio."
- p. 1 §2 Previous Work
Depoortere et al
- p. 8 Reference [2]
V. de Poortere
- p. 1 §2 Previous Work
Papageorgiou et al [18]
- p. 8 Reference [18]
C. Papageorgiou and T. Poggio.
Judge's note: Both verified on the page 1 and page 8 images. Surname form differs between text and reference list, and 'et al' is used for a two-author entry. The cited works remain identifiable; a minor style inconsistency.
Second check: none (judge only)
Report a problem with this finding
Typo Copyediting slips: "if they are much smaller that the local" (than); Fig. 6 caption "(e) It’s computed R-HOG descriptor." (Its); Acknowledgments "We thanks Cordelia Schmid" (thank).
- p. 2 §3, paragraph beginning 'The HOG/SIFT representation'
if they are much smaller that the local
- p. 8 Fig. 6 caption
(e) It’s computed R-HOG descriptor.
- p. 8 Acknowledgments
We thanks Cordelia Schmid
- p. 2 Sec 3
much smaller that the local
Judge's note: All three are visible on the page images (page 2 for 'smaller that'; page 8 for 'It's computed' and 'We thanks'). Plain grammar/spelling errors with evident fixes.
Second check: none (judge only)
Report a problem with this finding
Typo Text says "Fig. 3(a) also shows MIT’s best parts based and monolithic detectors" but Figure 3 has no (a)/(b) panel labels; its caption identifies panels as "(left) MIT and (right) INRIA".
- p. 3 §5 Results
Fig. 3(a) also shows MIT’s
- p. 4 Figure 3 caption
on (left) MIT and (right) INRIA data sets
Judge's note: Page 3 image shows 'Fig. 3(a)'. Page 4 image shows Figure 3 with two unlabelled plots and a caption using (left)/(right). The target (the left, MIT plot) is recoverable, so this is only a cosmetic label inconsistency.
Second check: none (judge only)
Report a problem with this finding
Typo Fig. 3 (left, MIT) legend reads "Lin. G−ShaceC" and "Lin. E−ShaceC"; the right panel legend and the text use 'ShapeC'.
- p. 4 Figure 3, left (MIT) plot legend
Lin. G−ShaceC
- p. 4 Figure 3, left (MIT) plot legend
Lin. E−ShaceC
- p. 4 Figure 3, right panel legend
Lin. G−ShapeC
Judge's note: The crop and the page 4 image show 'ShaceC' in the left legend and 'G−ShapeC'/'E−ShapeC' in the right legend. The misspelling is in the figure itself, not an extraction artefact.
Second check: none (judge only)
Report a problem with this finding
Typo "using the Bhattacharya distance between them" — the usual spelling is 'Bhattacharyya'.
- p. 6 §6.4 Block Normalization schemes
using the Bhattacharya distance between them
- p. 6 Block normalization schemes
Bhattacharya distance
Only the authors can say what is right here.
Judge's note: Rests on outside knowledge: the standard spelling of the distance is 'Bhattacharyya'. The page 6 image shows 'Bhattacharya' as quoted. This variant transliteration is common in the literature; cosmetic only.
Second check: none (judge only)
Report a problem with this finding
Typo Acknowledgments (p. 8) print "We thanks Cordelia Schmid for many useful comments."; the verb should be "thank".
- p. 8 Acknowledgments
We thanks Cordelia Schmid
- p. 8
We thanks Cordelia Schmid for many useful comments.
Judge's note: Read the crop of page 8: the Acknowledgments read "We" at the end of one line and "thanks Cordelia Schmid" at the start of the next. The subject-verb agreement error is in the PDF itself, not an extraction artefact.
Second check: none (judge only)
Report a problem with this finding
Typo Figure 6 caption (p. 8) prints "(e) It’s computed R-HOG descriptor."; the possessive "Its" is meant.
- p. 8 Figure 6 caption, item (e)
It’s computed R-HOG descriptor
- p. 8 Figure 6 caption
(e) It’s computed R-HOG descriptor.
Judge's note: Read the crop of the Figure 6 caption: the apostrophe is visibly printed in "It’s computed R-HOG descriptor", following "(d) A test image." The sense is possessive (the test image's descriptor), so the contraction is a typo.
Second check: none (judge only)
Report a problem with this finding
Typo Reference [10] (p. 8) prints the editor's name as "B. Schlkopf, C. Burges, and A. Smola, editors"; the name appears to have lost a letter (Schölkopf).
- p. 8 Reference [10]
B. Schlkopf, C. Burges, and A. Smola
Only the authors can say what is right here.
Judge's note: Rests on outside knowledge: the editor of Advances in Kernel Methods - Support Vector Learning is B. Schölkopf. The page image does print "Schlkopf" with no "ö" or "o", so the dropped character is in the PDF rendering, not only in the text layer. The correct spelling cannot be verified from the files.
Second check: none (judge only)