@hlee presents the 2D sensor module, how it works, and what it can do. This new SM allows Monty to deal with certain types of distortions. For example, it can now learn a logo on any surface (e.g. on a flat surface) and then recognize it on any other surface (e.g. on a round mug).
Read more about this new feature here:
Summary Video
Main Video
00:00 Introduction
00:39 Monty’s New Capability: Learning Surface Models
02:16 Motivation: Recognizing Logos on Objects for Compositional Models
07:52 Results: Surface (2D) Models Allow for Generalization Across Different Surfaces
13:29 Methods: Detecting the Dominant Edge in a Patch
41:27 Discussion: V1 Receptive Fields Beyond Orientations
56:06 Discussion: Shadows and Edges
01:05:45 Discussion: An Example of the Luminance Discounting Problem
01:08:13 Methods: How the 2D Reference Frame Was Learned (Tangents and a Moving Reference Frame)
01:09:58 What Kind of Surfaces Can be Unrolled?
01:12:40 Summary & Special Mentions
Get the code: GitHub - thousandbrainsproject/tbp.monty: Monty is a sensorimotor learning framework based on the thousand brains theory of the neocortex. · GitHub
Chat with our community: https://thousandbrains.discourse.group
Read the docs: https://thousandbrainsproject.readme.io
Follow us on Bluesky: Bluesky
Follow us on LinkedIn: Thousand Brains Project | LinkedIn
Follow us on X: Thousand Brains Project (@1000brainsproj) / X
Join our meetup group: Thousand Brains Project | Meetup
Join our mailing list: https://thousandbrains.org
See our future work roadmap: Project Roadmap