Development in Object Detection - Oregon State University
Transcript of Development in Object Detection - Oregon State University
![Page 1: Development in Object Detection - Oregon State University](https://reader035.fdocuments.us/reader035/viewer/2022071602/613d521c736caf36b75bef72/html5/thumbnails/1.jpg)
Development in Object Detection
Junyuan Lin May 4th
![Page 2: Development in Object Detection - Oregon State University](https://reader035.fdocuments.us/reader035/viewer/2022071602/613d521c736caf36b75bef72/html5/thumbnails/2.jpg)
Line of Research
[1] N. Dalal and B. Triggs. Histograms of oriented gradients for human detection, CVPR 2005.
[2] P. Felzenszwalb, D. McAllester, and D. Ramanan. A discriminatively trained, multiscale, deformable part model, CVPR 2008.
[3] C. Desai, D. Ramanan, C. Fowlkes. Discriminative Models for Multi-Class Object Layout, ICCV 2009.
HOG Feature template
Deformable Part Model
Multi-class Object Relationship
![Page 3: Development in Object Detection - Oregon State University](https://reader035.fdocuments.us/reader035/viewer/2022071602/613d521c736caf36b75bef72/html5/thumbnails/3.jpg)
HOG Feature template
Deformable Part Model
Multi-class Object Relationship
Linear SVM
Latent SVM
Structural SVM
Y structured output
Max-margin Formulation
![Page 4: Development in Object Detection - Oregon State University](https://reader035.fdocuments.us/reader035/viewer/2022071602/613d521c736caf36b75bef72/html5/thumbnails/4.jpg)
•Concatenate HOG Features in a search window into feature vector•Formulate the object detection as a binary linear classifier problem
•Sliding window scanning over the HOG feature pyramid •Output location with highest score f(x)•Possible post process (Non-maxima suppression)
Dalal & Triggs Detector
![Page 5: Development in Object Detection - Oregon State University](https://reader035.fdocuments.us/reader035/viewer/2022071602/613d521c736caf36b75bef72/html5/thumbnails/5.jpg)
Linear SVM classifier
![Page 6: Development in Object Detection - Oregon State University](https://reader035.fdocuments.us/reader035/viewer/2022071602/613d521c736caf36b75bef72/html5/thumbnails/6.jpg)
Learned weighted Filter
Training sample Positive weights Negative weights
*
![Page 7: Development in Object Detection - Oregon State University](https://reader035.fdocuments.us/reader035/viewer/2022071602/613d521c736caf36b75bef72/html5/thumbnails/7.jpg)
Deformable Part Model
Multiscale model captures features at two-resolution
relative to p0
Slide by P. Felzenszwalb
![Page 8: Development in Object Detection - Oregon State University](https://reader035.fdocuments.us/reader035/viewer/2022071602/613d521c736caf36b75bef72/html5/thumbnails/8.jpg)
Score of a hypothesis
Latent variable
relative to anchor position
Slide by P. Felzenszwalb
![Page 9: Development in Object Detection - Oregon State University](https://reader035.fdocuments.us/reader035/viewer/2022071602/613d521c736caf36b75bef72/html5/thumbnails/9.jpg)
Latent SVM Training
Hinge loss
Learn
Not convex in general !
Semi-convexity:
is concave for positive examples
is convex for negative examples
For positive examples make convex by fixing the latent variable for each positive training example.
![Page 10: Development in Object Detection - Oregon State University](https://reader035.fdocuments.us/reader035/viewer/2022071602/613d521c736caf36b75bef72/html5/thumbnails/10.jpg)
Latent SVM Training cont’d
Coordinate descent optimization approach:Initialize and iterate:
•Fix pick best for each positive example.
•Fix optimize over by quadratic programming or gradient descent
Model training procedure:
-Root Filter : train an initial root filter F0 using linear SVM without latent variable.-Part Filter : Initialize six parts from root filter F0 with the same shape,
place at anchor positions with most positive energy in F0 . The size of each part filter is determined to take 80% of the area of F0.
•Initialization:
•Train the latent SVM model
![Page 11: Development in Object Detection - Oregon State University](https://reader035.fdocuments.us/reader035/viewer/2022071602/613d521c736caf36b75bef72/html5/thumbnails/11.jpg)
Example Results
Slide by P. Felzenszwalb
![Page 12: Development in Object Detection - Oregon State University](https://reader035.fdocuments.us/reader035/viewer/2022071602/613d521c736caf36b75bef72/html5/thumbnails/12.jpg)
Quantitative Result
![Page 13: Development in Object Detection - Oregon State University](https://reader035.fdocuments.us/reader035/viewer/2022071602/613d521c736caf36b75bef72/html5/thumbnails/13.jpg)
Multi-class Object LayoutPascal Dataset: Multiple objects of multi-class in a single image. Problems with traditional binary classification + sliding window approach:• Intra-class: Require ad-hoc non-maxima suppression as post processing.• Inter-class: multiple class models are searched independently over images.
Heuristically forcing mutual exclusion.
Need a principal way to model spatial statistics between objects
![Page 14: Development in Object Detection - Oregon State University](https://reader035.fdocuments.us/reader035/viewer/2022071602/613d521c736caf36b75bef72/html5/thumbnails/14.jpg)
FormulationFormulate as a structured prediction problem.
X: a collection of overlapping windows at various scales covering the entire imageY: the entire label vector for the set of all windows in an image.
spatial relationship between prediction
![Page 15: Development in Object Detection - Oregon State University](https://reader035.fdocuments.us/reader035/viewer/2022071602/613d521c736caf36b75bef72/html5/thumbnails/15.jpg)
Spatial context descriptor
Output score of local detectors
The parameter learning and inference can be formulated as a structural SVM framework.
![Page 16: Development in Object Detection - Oregon State University](https://reader035.fdocuments.us/reader035/viewer/2022071602/613d521c736caf36b75bef72/html5/thumbnails/16.jpg)
Structural SVM
Joint feature map
•Exponential Parameters•Exponential Constraints
represent Y using features extract from X,Y
Learn weights so that is max for correct y
Multi-class SVM
![Page 17: Development in Object Detection - Oregon State University](https://reader035.fdocuments.us/reader035/viewer/2022071602/613d521c736caf36b75bef72/html5/thumbnails/17.jpg)
Structural SVM cont’dn-Slack Formulation: (margin rescaling)
1-Slack Formulation:
Slide by T. Joachims
![Page 18: Development in Object Detection - Oregon State University](https://reader035.fdocuments.us/reader035/viewer/2022071602/613d521c736caf36b75bef72/html5/thumbnails/18.jpg)
Cutting-Plane Algorithm• Input: •• REPEAT
– FOR– Compute
– ENDFOR– IF
– optimize StructSVM over– ENDIF
• UNTIL has not changed during iteration
_
[Jo06] [JoFinYu08]
Add constraint to working set
Find most violated
constraint
Violated by more than ε ?
![Page 19: Development in Object Detection - Oregon State University](https://reader035.fdocuments.us/reader035/viewer/2022071602/613d521c736caf36b75bef72/html5/thumbnails/19.jpg)
Theoretical BoundTheorem: Given any , the number of constraints added to working set
S is bounded by
where
so that is a feasible solution.
• Number of constraints is independent of training examples.• Linear time training algorithm.
+
∆
εCR
CR
2
2216
4log
![Page 20: Development in Object Detection - Oregon State University](https://reader035.fdocuments.us/reader035/viewer/2022071602/613d521c736caf36b75bef72/html5/thumbnails/20.jpg)
Summary of Structural SVM• Training: (cutting plane algorithm)
• Prediction:
• Application Specific Design of Model– Loss function:– Representation: joint feature map– Algorithm to compute:
![Page 21: Development in Object Detection - Oregon State University](https://reader035.fdocuments.us/reader035/viewer/2022071602/613d521c736caf36b75bef72/html5/thumbnails/21.jpg)
Component Formulation
0-1 loss functionjoint feature map
Greedy forward search algorithm
![Page 22: Development in Object Detection - Oregon State University](https://reader035.fdocuments.us/reader035/viewer/2022071602/613d521c736caf36b75bef72/html5/thumbnails/22.jpg)
Slide by D. Ramanan
![Page 23: Development in Object Detection - Oregon State University](https://reader035.fdocuments.us/reader035/viewer/2022071602/613d521c736caf36b75bef72/html5/thumbnails/23.jpg)
Slide by D. Ramanan
![Page 24: Development in Object Detection - Oregon State University](https://reader035.fdocuments.us/reader035/viewer/2022071602/613d521c736caf36b75bef72/html5/thumbnails/24.jpg)
Slide by D. Ramanan
![Page 25: Development in Object Detection - Oregon State University](https://reader035.fdocuments.us/reader035/viewer/2022071602/613d521c736caf36b75bef72/html5/thumbnails/25.jpg)
Slide by D. Ramanan