View page as slide show

Demos and presentations

Defomable detector

deformable_detections.jpg

Improving Cascade of Classifiers by Sliding Window Alignment

Fast Learnable Object Tracking and Detection

Fast Video Browsing

  • off-line video processing: cca 3x faster than the duration of the video sequence
  • detection and tracking of objects in video using learning methods
  • single training image for learning of SURF detector and Linear Predictor tracker
  • reinforcement learning: the system selects only the useful additional training images and the user decides whether to use them or not in the incremental learning of the tracker for another browsing iteration.
  • usually 2-3 iterations are sufficient

Automatic Video Cut

 Picture

automatic_cut Video consists of two parts. First is a video cut of one episode of the Fawlty Towers Series, where the shots of one room with the picture object are present. The second part demonstrates the power of incremental learning where the predictor tracks a human face with changing appearance.

 
misc/projects/advanced_predictors/demos/start.txt · Last modified: 2013/10/04 13:02 (external edit)