|
Deep Learning in Image Forensics (DLinIF)10 ECTS (englische Bezeichnung: Deep Learning in Image Forensics)
Modulverantwortliche/r: Christian Riess, Marc Stamminger Lehrende:
Luisa Verdoliva, Christian Riess
Startsemester: |
SS 2018 | Dauer: |
1 Semester | Turnus: |
unregelmäßig |
Präsenzzeit: |
30 Std. | Eigenstudium: |
270 Std. | Sprache: |
Englisch |
Lehrveranstaltungen:
Empfohlene Voraussetzungen:
Participants of this class are expected to bring a working knowledge in python. Also, participants are expected to bring theoretical and practical knowledge from classes in either pattern recognition, machine learning, or computer graphics.
Inhalt:
Is an image pristine, or has its content been edited? Manipulation detection is one of the goals in image forensics. In this hands-on class, we will look at the currently most popular learning-based approaches to manipulation detection. A particular focus of this class will lie on the task of training a deep neural network for image manipulation detection. The topic will be introduced in a few lectures. Then, the participants will experiment with an own implementation of a neural network for manipulation detection. The network training and performance assessment will be done on provided benchmark data. In the course of the semester, weaknesses in the network performance will be analyzed, and based on this analysis, the network will be gradually improved.
Tentative semester outline:
weeks of May 7 until May 18: introductory lectures on image forensics and the tensorflow framework.
weeks of May 21 until June 22: introductory project assignment
weeks of June 25 until July 6: advanced lectures on deep learning in image forensics
July 9 until September 10: main project on image manipulation detection using deep learning
Lernziele und Kompetenzen:
- Anwenden
- Participants implement and train an existing deep learning architecture
- Analysieren
- Participants explore weaknesses of the trained network on provided benchmark data
- Evaluieren (Beurteilen)
- Participants validate their solution using common quality criteria in the field of image forensics
- Erschaffen
- Participants create and implement strategies to improve their developed solution against selected failure cases.
Bemerkung:
Please register to this class via email to Christian Riess.
Studien-/Prüfungsleistungen:
Deep Learning in Image Forensics (Prüfungsnummer: 922926)
(englischer Titel: Deep Learning in Image Forensics)
- Prüfungsleistung, mehrteilige Prüfung, benotet, 10.0 ECTS
- Anteil an der Berechnung der Modulnote: 100.0 %
- weitere Erläuterungen:
The grade consists of: 50% implementation of software for the predefined task, 25% presentation of the software to the supervisors 25% compact description of the software and the results in a written text.
- Prüfungssprache: Englisch
- Erstablegung: SS 2018, 1. Wdh.: WS 2018/2019
1. Prüfer: | Christian Riess, | 2. Prüfer: | Marc Stamminger |
|
|
|
|
UnivIS ist ein Produkt der Config eG, Buckenhof |
|
|