face detection - cse.wustl.edu · pdf filechallenges of face detection • sliding window...
TRANSCRIPT
![Page 1: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/1.jpg)
Face Detection
The “Margaret Thatcher Illusion”, by Peter Thompson
![Page 2: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/2.jpg)
Funny Nikon ads"The Nikon S60 detects up to 12 faces."
![Page 3: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/3.jpg)
Funny Nikon ads"The Nikon S60 detects up to 12 faces."
![Page 4: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/4.jpg)
Consumer application: Apple iPhotoThings iPhoto thinks are faces
![Page 5: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/5.jpg)
Applica'ons:Computa'onalphotography
![Page 6: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/6.jpg)
Consumer application: Apple iPhoto
http://www.apple.com/ilife/iphoto/
![Page 7: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/7.jpg)
Consumer application: Apple iPhotoCan be trained to detect pets!
http://www.maclife.com/article/news/iphotos_faces_recognizes_cats
![Page 8: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/8.jpg)
8
![Page 9: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/9.jpg)
9
![Page 10: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/10.jpg)
Challenges of face detection• Sliding window detector must evaluate tens of
thousands of location/scale combinations • Faces are rare: 0–10 per image
• Should make quick decisions at non-face regions • A megapixel image has ~106 pixels and a comparable number of
candidate face locations • To avoid having a false positive in every image image, our false
positive rate has to be less than 10-6
![Page 11: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/11.jpg)
One simple method: skin detection
Skin pixels have a distinctive range of colors • Corresponds to region(s) in RGB color space
– for visualization, only R and G components are shown above
skin
Skin classifier • A pixel X = (R,G,B) is skin if it is in the skin region• But how to find this region?
![Page 12: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/12.jpg)
Skin detection
Learn the skin region from examples • Manually label pixels in one or more “training images” as skin or not skin • Plot the training data in RGB space
– skin pixels shown in orange, non-skin pixels shown in blue – some skin pixels may be outside the region, non-skin pixels inside. Why?
Skin classifier • Given X = (R,G,B): how to determine if it is skin or not?
![Page 13: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/13.jpg)
Skin classification techniques
Skin classifier • Given X = (R,G,B): how to determine if it is skin or not? • Nearest neighbor
– find labeled pixel closest to X – choose the label for that pixel
• Data modeling – fit a model (curve, surface, or volume) to each class
• Probabilistic data modeling – fit a probability model to each class
![Page 14: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/14.jpg)
Skin detection results
![Page 15: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/15.jpg)
The Viola/Jones Face Detector• A seminal approach to real-time object detection • Training is slow, but detection is very fast • Key ideas
• Integral images for fast feature evaluation • Boosting for feature selection • Attentional cascade for fast rejection of non-face windows
P. Viola and M. Jones. Rapid object detection using a boosted cascade of simple features. CVPR 2001.
P. Viola and M. Jones. Robust real-time face detection. IJCV 57(2), 2004.
![Page 16: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/16.jpg)
Image Features
“Rectangle filters”
Value =
∑ (pixels in white area) – ∑ (pixels in black area)
![Page 17: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/17.jpg)
Example
Source
Result
![Page 18: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/18.jpg)
Example
Source
Result
![Page 19: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/19.jpg)
Fast computation with integral images• The integral image
pixel (x,y) stores the sum of the values above and left
• This can quickly be computed in one pass through the image
(x,y)
![Page 20: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/20.jpg)
5
Integral imagesWhat’s the sum of pixels in the blue rectangle?
input image integral image
A B
C D
![Page 21: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/21.jpg)
6
Integral images
integral image
A B
C D
![Page 22: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/22.jpg)
7
Integral images
integral image
A
![Page 23: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/23.jpg)
8
Integral images
integral image
B
![Page 24: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/24.jpg)
9
Integral images
integral image
C
![Page 25: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/25.jpg)
10
Integral images
integral image
D
![Page 26: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/26.jpg)
Integral imagesWhat’s the sum of pixels in the rectangle?
DA CB
![Page 27: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/27.jpg)
Computing sum within a rectangle
• Let A,B,C,D be the values of the integral image at the corners of a rectangle
• Then the sum of original image values within the rectangle can be computed as:
sum = A – B – C + D • Only 3 additions are
required for any size of rectangle!
D B
C A
Lana Lazebnik
![Page 28: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/28.jpg)
Computing a rectangle feature
-1 +1+2-1
-2+1
Integral Image
![Page 29: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/29.jpg)
Feature selection• For a 24x24 detection region, the number of
possible rectangle features is ~160,000!
![Page 30: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/30.jpg)
Feature selection• For a 24x24 detection region, the number of
possible rectangle features is ~160,000! • At test time, it is impractical to evaluate the entire
feature set • Can we create a good classifier using just a small
subset of all possible features? • How to select such a subset?
![Page 31: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/31.jpg)
Boosting• Boosting is a classification scheme that combines weak
learners into a more accurate ensemble classifier • Weak learners based on rectangle filters:
window
value of rectangle feature
threshold
![Page 32: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/32.jpg)
Boosting• Boosting is a classification scheme that combines weak
learners into a more accurate ensemble classifier • Weak learners based on rectangle filters:
• Ensemble classification function:window
value of rectangle feature
threshold
learned weights
![Page 33: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/33.jpg)
Boosting
Weak Classifier 1
Slide credit: Paul Viola
![Page 34: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/34.jpg)
Boosting
Weights Increased
![Page 35: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/35.jpg)
Boosting
Weak Classifier 2
![Page 36: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/36.jpg)
Boosting
Weights Increased
![Page 37: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/37.jpg)
Boosting
Weak Classifier 3
![Page 38: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/38.jpg)
Boosting
Final classifier is a combination of weak classifiers
![Page 39: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/39.jpg)
Boosting: training
• Initially, weight each training example equally
• In each boosting round:– find the best weak classifier– raise weights of misclassified training examples
• Final classifier is linear combination of all weak classifiers– weight of each learner is directly proportional to its accuracy)
• Exact formulas for re-weighting and combining weak classifiers depend on the particular boosting scheme
Slide credit: Lana Lazebnik
![Page 40: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/40.jpg)
Boosting for face detection• First two features selected by boosting:
• This feature combination can yield 100% detection rate and 50% false positive rate
![Page 41: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/41.jpg)
Boosting for face detection• A 200-feature classifier can yield 95% detection rate
and a false positive rate of 1 in 14084
Receiver operating characteristic (ROC) curve
![Page 42: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/42.jpg)
• Even if the filters are fast to compute, each new image has a lot of possible windows to search.
• How to make the detection more efficient?
![Page 43: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/43.jpg)
Cascading classifiers for detection
Kristen Grauman
![Page 44: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/44.jpg)
Cascading classifiers for detection
• Form a cascade with low false negative rates early on• Apply less accurate but faster classifiers first to
immediately discard windows that clearly appear to be negative Kristen Grauman
![Page 45: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/45.jpg)
Attentional cascade• Start with simple classifiers which reject many
negatives but do not miss positives • Positive response moves to the next • Negative response means immediate rejection
FACEIMAGE SUB-WINDOW Classifier 1
TClassifier 3
T
F
NON-FACE
TClassifier 2
T
F
NON-FACE
F
NON-FACE
![Page 46: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/46.jpg)
Attentional cascade• Chain classifiers that are
progressively more complex and have lower false positive rates:
vs false neg determined by
% False Pos
% D
etec
tion
0 50
0
100
FACEIMAGE SUB-WINDOW Classifier 1
TClassifier 3
T
F
NON-FACE
TClassifier 2
T
F
NON-FACE
F
NON-FACE
Receiver operating characteristic
![Page 47: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/47.jpg)
Training the cascade• Set target detection and false positive rates for
each stage • For each stage
• Keep adding features to achieve the target detection • Lower AdaBoost threshold to maximize detection
(as opposed to minimizing total classification error) • Test on a validation set • If false positive rate is not low enough, then add another stage
• Note: Use false positives from current stage as the negative training examples for the next stage
FACEIMAGE SUB-WINDOW Classifier 1
TClassifier 3
T
F
NON-FACE
TClassifier 2
T
F
NON-FACE
F
NON-FACE
![Page 48: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/48.jpg)
The implemented system
• Training Data • 5000 faces
– All frontal, rescaled to 24x24 pixels
• 300 million non-faces – 9500 non-face images
• Faces are normalized – Scale, translation
• Many variations • Across individuals • Illumination • Pose
![Page 49: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/49.jpg)
System performance• Training time: “weeks” on 466 MHz Sun
workstation • 38 layers, total of 6061 features • Average of 10 features evaluated per window on
test set • “On a 700 Mhz Pentium III processor, the face
detector can process a 384 by 288 pixel image in about .067 seconds” • 15 Hz • 15 times faster than previous detector of comparable accuracy
(Rowley et al., 1998)
![Page 50: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/50.jpg)
Output of Face Detector on Test Images
![Page 51: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/51.jpg)
Other detection tasks
Facial Feature Localization
Male vs. female
Profile Detection
![Page 52: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/52.jpg)
Profile Detection
![Page 53: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/53.jpg)
Profile Features
![Page 54: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/54.jpg)
Viola-Jones detector: summary
•Train with 5K positives, 350M negatives•Real-time detector using 38 layer cascade•6061 features in all layers•[Implementation available in OpenCV: http://www.intel.com/technology/computing/opencv/]
Faces
Non-faces
Train cascade of classifiers with
AdaBoost
Selected features, thresholds, and weights
New image
Appl
y to
eac
h
subw
indo
w
Kristen Grauman
![Page 55: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/55.jpg)
Application: streetview
![Page 56: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/56.jpg)
3434
Application: streetview
![Page 57: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/57.jpg)
3535
Application: streetview
![Page 58: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/58.jpg)
What other categories are amenable to window-based representation?
![Page 59: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/59.jpg)
Perc
eptu
al a
nd S
enso
ry A
ugm
ente
d Co
mpu
ting
Visu
al O
bjec
t Re
cogn
itio
n Tu
tori
alVi
sual
Obj
ect
Reco
gnit
ion
Tuto
rial
Pedestrian detection• Detecting upright, walking humans also possible using sliding window’s
appearance/texture; e.g.,
SVM with HoG [Dalal & Triggs, CVPR 2005]
Kristen Grauman
![Page 60: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/60.jpg)
Perc
eptu
al a
nd S
enso
ry A
ugm
ente
d Co
mpu
ting
Visu
al O
bjec
t Re
cogn
itio
n Tu
tori
alVi
sual
Obj
ect
Reco
gnit
ion
Tuto
rial
Limitations (continued)
•Not all objects are “box” shaped
Kristen Grauman
![Page 61: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/61.jpg)
Perc
eptu
al a
nd S
enso
ry A
ugm
ente
d Co
mpu
ting
Visu
al O
bjec
t Re
cogn
itio
n Tu
tori
alVi
sual
Obj
ect
Reco
gnit
ion
Tuto
rial
Limitations (continued)
•Non-rigid, deformable objects not captured well with representations assuming a fixed 2d structure; or must assume fixed viewpoint
•Objects with less-regular textures not captured well with holistic appearance-based descriptions
Kristen Grauman
![Page 62: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/62.jpg)
Perc
eptu
al a
nd S
enso
ry A
ugm
ente
d Co
mpu
ting
Visu
al O
bjec
t Re
cogn
itio
n Tu
tori
alVi
sual
Obj
ect
Reco
gnit
ion
Tuto
rial
Limitations (continued)
•If considering windows in isolation, context is lost
Figure credit: Derek Hoiem
Sliding window Detector’s view
Kristen Grauman
![Page 63: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/63.jpg)
Perc
eptu
al a
nd S
enso
ry A
ugm
ente
d Co
mpu
ting
Visu
al O
bjec
t Re
cogn
itio
n Tu
tori
alVi
sual
Obj
ect
Reco
gnit
ion
Tuto
rial
Limitations (continued)
•In practice, often entails large, cropped training set (expensive)
•Requiring good match to a global appearance description can lead to sensitivity to partial occlusions
Image credit: Adam, Rivlin, & Shimshoni Kristen Grauman
![Page 64: Face Detection - cse.wustl.edu · PDF fileChallenges of face detection • Sliding window detector must evaluate tens of thousands of location/scale combinations • Faces are rare:](https://reader035.vdocument.in/reader035/viewer/2022062907/5a9daf867f8b9aee528b5b1e/html5/thumbnails/64.jpg)
Deep Learning
64