This course aims to teach learners about object recognition through concepts such as Bag of Words, pattern recognition in patches, forming a dictionary of words, and using codebooks for histogram similarity. The course covers skills like sampling patches and images, creating concept-specific codebooks, and understanding visual synonyms. The teaching method includes lectures on various topics related to object recognition. The course is intended for individuals interested in computer vision, image processing, and machine learning.
Overview
Syllabus
Object recognition
Bag of Words
Bags of Words
Capture the pattern in patch
The dimensionality
Sample many patches
Two types of sampling
Sample many images
Include all relevant variations
Form a dictionary of words
Ideally words cover similar patches
Count words per image
Codeboods
Learn histogram similarity
Similarity between two histograms
Classify unknown image
Concept-specific codebooks
Conclusion on concept-codebooks
Soft word assignment
Fisher vector
Conclusion on words
Codebook synonyms
How close are synonyms?
90% removed, same result
Visual synonym examples
Conclusion on visual synonyms
Convex reduced codebooks
Conclusion convex reduced
The where and what
What makes a boat a boat?
What is the object in the middle?
Where is evidence?
Where is evidence for an object?
The visual extent of an object
Context dominance
Object dominance
Object detail dominance
Pyramids: simple compositional
Exhaustive search
The need for high recall
The need for hierarchy
Selective search example
Selective search to get high recall
Average best overlap -88%
Classification with selective search
Conclusion on location
Two concepts in interaction
Taught by
UCF CRCV
