2013.04.10 Activity Recognition Meeting Report

• Meeting April 10, with AC, RM, Pierre Luc, Samira and Atousa

Short-term goals:

• Samira will work on improving her Keypoint detection for the challenge data. Specifically, she will investigate normalizing the aspect ratio.
  
• Pierre Luc will work on improving his Keypoint detection method and applying it first to Yoshua's class dataset, and then to the challenge data.
  
• Atousa will apply the activity recognition pipeline of Wang et al (state-of-the-art on Hollywood 2) to the problem of emotion recognition. There was a recognition that some of the hyperparameters (e.g. the length of the trajectories) will likely need to be adjusted.
• Roland will work with his student on his approach to activity recognition, to compare with Atousa's implementation of emotion recognition.