Scott gives a presentation on his work detecting saliency and using this information to help Monty move more efficiently. We discuss issues that arise without inhibition of return, as well as how this method could be modified to work well on compositional objects and more complex scenes.
Summary Video
Main Video
00:00 Intro
00:07 Salience and Segmentation with Compositional Objects
01:28 Limitations of just using Salience Maps
05:53 Is the Inhibition of Return Calculated in 2D or 3D Space?
14:27 Model Free Segmentation or Model-based?
19:11 Using Depth to Force Attention
28:00 What Monty sees - A Visualization
35:49 Other Work on Model-Free Segmentation and Saliency Detection
43:45 Further Discussion on Model-Free vs. Model-Based Way to Define Attended Regions
