CODEWITHSUNDEEP
c++opencvcomputer-visionfeature-extraction
SneakyCrab
SneakyCrab

Reputation: 33

size of HOG feature dimension

I am confused when trying to understanding the size of HOG feature in this paper.Mid-level Elements for Object Detection..In Dalal's work,a 64*128 pixel window is represented by a 4*9*7*15=3780 dimension HOG feature.But in these paper a 64*64 pixel window is represented by a 6*6*31 HOG feature.I have read the cited reference in it and still don't understand how it is calculated.Can anyone help me to understand the idea to extract 6*6*31 HOG feature per 64*64 window?What's the cellsize ,blocksize and stride in this paper？

Upvotes: 3

Views: 945

Answers (1)

Bashar Haddad
Bashar Haddad

Reputation: 368

You need to know that there are two different HOG features. There is the original one suggested by Dalal (where the images is divided into blocks and the block is divided into cells, exactly as you explained in the question). The second is the UoCTTI HoG feature.

to explain what you are asking about, let me start by explaining the number 31. This is the dimensions of the UoCTTI Hog feature described by this paper (P. F. Felzenszwalb, R. B. Grishick, D. McAllester, and D. Ramanan. Object detection with discriminatively trained part based models. PAMI, 2009). In this paper they use PCA to reduce the dimensions from 36 to 13, then they perform different types of augmentation to add the rest (18). 6x6 is the division of the image into cells.

Upvotes: 1

Related Questions