150 likes | 261 Vues
In Week 5, I conducted a thorough analysis using the PubFig83 dataset to run tests based on training and testing image IDs. I computed pairwise cosine distances between images and created a zero matrix to visualize relationships among them. By implementing a kNN graph and identifying top 5 neighbors, I refined the dataset for accurate modeling. I also reviewed a significant research paper outlining recent advancements in face detection algorithms, focusing on the Viola-Jones detector and boosted algorithms such as AdaBoost. Moving forward, I plan to refine my code and consult with my mentor for further direction.
E N D
Week 5 Report Shelby Thompson
This week… • 1 research paper • Coded more • Used PubFig83 dataset to run tests on
Using the PubFig83 dataset… • Used training, testing and image IDs • Compute pairwise distance between images • Did so by computing cosine distance between matrix rows
Using PubFig83 dataset… • Next created a zero matrix as big as distance matrix • Iterated over the training and testing IDs • If the train and test IDs were equal set that spot in the zero matrix to 1
Created kNN graph… • Iterated over the distance matrix • Sorted it in declining order • Found top 5 k-neighboors • Set all other rows in the matrix to 0
Created kNN graph (cont)… • Iterated over each row/threshold • Iterate based on (i,i) where I is row number • Set all rows above threshold to 0 or Inf (they appear different)
Research Paper A Survey of Recent Advances in Face Detection By: Cha Zhang and Zhengyou Zhang (Briefly reviewed this article)
Article Focus… • Review past decade of development in face detection • Uses Viola-Jones face detector • Survey other algorithms and feature extractors • Goal is to find the best algorithm
Experiment… • The authors focused more on surveying and researching current face detection methods • Examined different face detection methods and how they improved • Focus on boosted algorithms, like AdaBoost • Examined popular programs like iPhoto and Picassa
Goals for next week… • Continue to work on my codes • Experiment with writing some of my MatLab code in Java • Find out what Enrique wants me to do next