DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING
UNIVERSITY OF CALIFORNIA, SAN DIEGO

• Implement one of Serge's suggested projects for CSE190a
• Zero Shot Learning
• Train computer vision classifiers to detect objects using no training images and only a text description
• Could implement the Lampert et.al CVPR'09 paper Learning To Detect Unseen Object Classes by Between-Class Attribute Transfer
• Use data from Animals With Attributes or Birds 200.
• Implement Human Pose Recognition Algorithm For the Kinect
• Could (partially) implement the Shotton et.al CVPR'11 paper Real-Time Human Pose Recognition in Parts from a Single Depth Image
• Fine-Grained Category Recognition
• Investigate which types of features are appropriate for classification of subordinate categories (e.g. recognizing bird species), where the differences between categories is very small
• Investigate to what extent localization information (e.g., knowing the location of object parts) improves recognition
• Could start with features and learning algorithms from VLFeat and VGG-MKL, or OpenCV
• Use images and part labels from Birds 200
• Tracking By Detection
• Try out a tracking by detection algorithm: Lepitit et al. Random Ferns, Babenko et al. MIL, Kalal et al. Tracking-Learning-Detection
• Possible project: can you use data captured/tracked from video to help train an object detector for novel images?
• Possible project: use tracker to track multiple body parts and create an interactive interface or game
• Implement one of these papers:
• Geometric min-Hashing: Finding a (Thick) Needle in a Haystack
• Scene Completion using Millions of Photographs
• Fast Ray Features for Learning Irregular Shapes
• Try to apply SIFT Flow to some application like face recognition
• Any other paper from the last 5 or so years from CVPR, ICCV, ECCV, or BMVC