Deep Learning for Computer Vision (IN2346) (2h + 2h, 6ECTS)
SS 2017, TU München
Lecture
MOODLE
We use Moodle for discussions and to distribute important information. Please check the News and Discussion boards regularly or subscribe to them.
NEW LOCATION AND SCHEDULE!
Due to the high demand for the course we changed the schedule from Tuesday to Friday to get a bigger lecture room.
Thursday (16:00-18:00) - Walter-Hieber-Hörsaal (Chemistry building)
Friday (14:00-16:00) - MI Hörsaal 2
Lecturers: Dr. Laura Leal-Taixé, Prof. Dr. Matthias Niessner
ECTS: 6
SWS: 4
Tutorial
Date: on Fridays
Tutor: Thomas Frerix, Tim Meinhardt
Content
- Lecture 1 (27.04): Introduction to Computer Vision and history of Deep Learning.
- Lecture 2 (28.04): Machine Learning basics 1: linear classification, maximum likelihood.
- Lecture 3 (04.05): Machine Learning basics 2: logistic regression, perceptron
- Lecture 4 (11.05): Introduction to neural networks and their optimization, SGD, Back-propagation.
- Lecture 5 (18.05): Training Neural Networks Part 1: regularization, activation functions, weight initialization, gradient flow, batch normalization, hyperparameter optimization.
- Lecture 6 (01.06): Training Neural Networks Part 2: parameter updates, ensembles, dropout.
- Lecture 7 (08.06): Convolutional Neural Networks.
- Lecture 8 (22.06): CNN for object detection (from MNIST to ImageNet), visualizing CNN (DeepDream).
- Lecture 9 (29.06): Prominent architectures: GoogleNet, ResNet.
- Lecture 10 (06.07): Generative Adversarial nets + Recurrent networks (NLP).
- Lecture 11 (13.07): LSTMs + Reinforcement Learning.
- Special lecture (20.07,27.07): to be announced.
Prerequisites
Passion for mathematics and the use of machine learning in order to solve complex computer vision problems. The course will be focused on practical projects, therefore, previous knowledge of a programming language, preferably Python , is desired.
Tentative exercise schedule
EXERCISE 1:
- Topics: Linear classifiers, multinomial regression, two-layer neural net.
- Starting date: May 5th
- Due date: May 17th
EXERCISE 2:
- Topics: Fully connected nets, dropout, batch normalization.
- Starting date: May 19th
- Due date: May 31th
EXERCISE 3:
- Topics: Convolutional neural networks, large-scale project with PyTorch.
- Starting date: June 2nd
- Due date: June 21st
FINAL PROJECT:
- Topic: each group will make a project proposal.
- Project proposal due date: June 28th
- Starting date: June 30th
- UPDATED!! Poster due date: Monday, August 7th, 23:59 (details in moodle news post)
- UPDATED!! Poster presentation: Thursday, August 10th
FINAL EXAM:
- Exam date: August 18th
For details about the exercises, please read the README, which is distributed with each exercise. In particular, you will submit your solutions via this website.
Lecture Slides
tba.
Contact us
If you have any questions regarding the organization of the course, do not hesitate to contact us at: dl4cv@vision.in.tum.de
For questions on the syllabus, exercises or any other questions on the content of the lecture, we will use the Moodle discussion board.
We offer to discuss questions in person in our office hours:
Dr. Laura Leal-Taixé: Wednesdays 1-2pm (room 02.09.044)
Prof. Dr. Matthias Niessner: Tuesdays 2-3pm (room 02.13.042)
Thomas Frerix: Thursdays 1-2pm (room 02.09.035)
Tim Meinhardt: Mondays 1-2pm (room 02.09.041)