Project Course: Computer Vision and Machine Learning

Instructors:	Christoph Lampert
Teaching Assistant:	Georg Martius
Meetings:	weekly on Tuesdays, 15:30 in Second Floor Seminar Room Lab Building East
Website:	official page
	Wiki (for exchanging information amongst participants)

announcements schedule references

Overview

This is a project course: students will work together in small teams to develop the Artificial Intelligence for remotely controlling a (toy) race car on a track. Instead of a final exam there will be a decisive race.

The course is open to interested students from all subject areas. However, solid programming skills and knowledge about algorithms are required. Knowledge about image processing and machine learning will be helpful, but can also be acquired over the course of the course.

Formalities:

Full semester course, 3 ECTS (one meeting per week + free team work)
please register in IQ

Announcements

github SDK contains new much easier code to control cars (Thanks Georg!)
Next meeting: Oct 20, LBE (=Lab Building East), 2nd floor seminar room
~~Next meeting: Oct 13, LBE (=Lab Building West), 2nd floor seminar room~~
Because of the large number of registrations, we'll offer the course again in Spring!
~~First meeting: Oct 09, Mondi 1~~

Materials

API to control cars: https://github.com/georgmartius/drive-sdk

car name	command (contains bluetooth ID)
boson	./vehicle-tool --device D9:81:41:5C:D4:31
katal	./vehicle-tool --device D8:64:85:29:01:C0
kourai	./vehicle-tool --device EB:0D:D8:05:CA:1A
rho	./vehicle-tool --device E6:D8:52:F1:D9:43
nuke	./vehicle-tool --device C5:34:5D:26:BE:53
hadion	./vehicle-tool --device D4:48:49:03:98:95

Example code how to access shared memory (C and Python) ZIP
Example camera image to try things with PPM PNG

References: Computer Vision

Documentation of the OpenCV library . For example, general image processing, background subtraction
A textbook about Computer Vision (contains a link to a draft PDF)

References: Machine Learning

A textbook about Reinforcement Learning (HTML). Printed copies will be available in the library soon.
Online material about Reinforcement Learning website video by David Silver (much more extensive than what you'll need)
Relevant keywords: e.g. Q-learning (also use in this recent Nature paper)

announcements schedule references