Download presentation
Presentation is loading. Please wait.
1
Week 6 University of Nevada, Reno
Emily Hand Week 6 University of Nevada, Reno
2
GML Adaboost Not Good Not moving forward with this approach The length of the feature vector must be less than the number of samples. Samples Features: HOG, LBP, 3DHistogram :( We tried different features, but nothing good came from it. Way too many false positives
3
False Positives
4
OpenTLD Detector Extracting their detector much more complicated than we thought. No documentation A lot of pre-compiled mex files Having issues with those
5
OpenTLD Detector Extracted their features :)
They resize each patch to [15 15] and then they reshape the patch to a column vector and normalize to zero mean unit variance (ZMUV). ZMUV Remove mean intensity value from an image and scale it with its variance
6
SVM Progress Histogram of Oriented Gradients (HOG)
Local Binary Pattern (LBP) 3D Histogram
7
Regular Histogram
8
More useful information
3D Histogram B G R 8x8x8 Bins More useful information
9
Some Results
10
Some Results
11
Next Week Motion Features Partial Occlusions Scale Changes
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.