Seminar: Current Trends in Deep Learning (IN2107, IN4515)
SS 2018, TU München
Seminar
Location: Room 02.09.023
Preliminary Meeting: 26.01.2018, 14:00 (please bring a device to access the internet)
Date: Thursday
Time: 14:00 - 16:00
Lecturer: Caner Hazirbas, Philip Haeusser
ECTS: 4
SWS: 2
The course will be held in English. Assignment to this course will be done in the matching system.
We will not favor/rank students who did not participate in the preliminary meeting.
Send your presentation via email to both advisors until 10.07.2018@23:59.
Description
Deep Learning studies computational methods that find patterns in structured data.
In this seminar we will discuss current trends in deep learning based on research articles.
Every student picks a recent research paper on deep learning for computer vision that she/he presents in the seminar.
Attendance is limited to 14 students (Master).
Registration: Matching system
Important Dates
Preliminary Meeting | 26.01.18, 2-4pm | Room 02.09.023 | Please bring a device to access the internet. |
First Lecture | 12.04.18, 2-4pm | Room 02.09.23 | Intro |
Second Lecture | 19.04.18, 2-4pm | Room 02.09.23 | Paper assignment |
Presentation | 10.07.18, 23:59pm | Presentation submission | |
Preset. Day 1 | 11.07.18, 10:15am-6pm | Room 02.09.23 | |
Preset. Day 2 | 12.07.18, 10:15am-6pm | Room 00.09.38 | |
Final Report | 22.07.18, 23:59pm | Report Submission |
Useful Documents
Preliminary meeting presentation Lecture lecture
Paper preference form
A LaTeX-template for your seminar report latex template
Paper Assigments
Student | Supervisor | Paper | Presentation date |
Felix Meissen | Philip | Multimodal Transfer: A Hierarchical Deep Convolutional Neural Network for Fast Artistic Style Transfer | 12.07.18 |
Alexander Ziller | Caner | Deformable Convolutional Networks | 11.07.18 |
Marius Obert | Philip | Attend in groups: a weakly-supervised deep learning framework for learning from web data | 12.07.18 |
Cem Yusuf Aydogdu | Caner | SSD: Single Shot MultiBox Detector | 11.07.18 |
Ayşe Hande Karatay | Philip | Look, Listen and Learn | 11.07.18 |
Felipe Peter | Caner | On-Demand Learning for Deep Image Restoration | 11.07.18 |
Yang An | Philip | DensePose: Dense Human Pose Estimation In The Wild | 11.07.18 |
Nail Ibrahimli | Caner | Unsupervised Learning of Depth and Ego-Motion from Video | 12.07.18 |
Marvin Alles | Philip | Towards 3D Human Pose Estimation in the Wild: a Weakly-supervised Approach | |
Maximilian Schneider | Caner | Learning to Track at 100 FPS with Deep Regression Networks | 11.07.18 |
Moritz Krügener | Philip | Fast Face-swap Using Convolutional Neural Networks | — |
Alexander Gaul | Caner | Temporal Convolutional Networks for Action Segmentation and Detection | 12.07.18 |
Furkan Mert Algan | Philip | Ambient sound provides supervision for visual learning | 11.07.18 |
Yehya Abouelnaga | Caner | Action-Decision Networks for Visual Tracking with Deep Reinforcement Learning | 12.07.18 |
Grading Rubric
We are going to grade you based on the following criteria:
Presentation (2/3)
- presentation has an accessible introduction
- presentation gets across the key idea of the paper
- presentation style: less text on slides, more figures
- presentation style: free but concise
- presentation: ability to answer questions
- active participation in discussion of other presentations
Report (1/3)
- report is well structured
- report reflects paper completely and correctly
- report is written in own words, no paraphrasing
- report contains secondary material
- report is spell-checked and written in proper English