ETRO VUB
About ETRO  |  News  |  Events  |  Vacancies  |  Contact  
Home Research Education Industry Publications About ETRO

Master theses

Current and past ideas and concepts for Master Theses.

Explainable deep learning for video recognition

Subject

Deep neural networks achieve state-of-the-art performance in many signal processing tasks, including video processing (denoising, super-resolution, reconstruction, …) and video analysis tasks (classification, anomaly detection, …). However, because of their high complexity, deep networks are often considered as black-box models, making it hard to assess what the model has learned, and explaining the decisions made by the model in human-understandable terms.

One method to explain model decisions consists in producing an “input saliency map”. It consists in visualizing the regions (or pixels) of the input that are considered decisive to explain the output, in the form of heatmaps. Such methods can be applied to video in to highlight the decisive visual features of the input frames.

Kind of work

The student will investigate existing visual explanations techniques and use them to explain decisions made by deep neural networks on a given video task (e.g., action classification on the UCF-101). Advantages and disadvantages of existing techniques will be discussed. We will then focus on implementing new explanation techniques and apply them on novel deep neural networks.

Framework of the Thesis

Kalfaoglu, M. Esat, Sinan Kalkan, and A. Aydin Alatan. "Late temporal modeling in 3d cnn architectures with bert for action recognition." European Conference on Computer Vision. Springer, Cham, 2020.
Montavon, Grégoire, et al. "Explaining nonlinear classification decisions with deep taylor decomposition." Pattern Recognition 65 (2017): 211-222.

Number of Students

1 or 2 students

Expected Student Profile

The student should have a background in linear algebra, machine learning, image processing and python programming.

Knowledge of deep learning and libraries like PyTorch, Keras or TensorFlow is a plus.

Promotor

Prof. Dr. Ir. Nikolaos Deligiannis

+32 (0)2 629 1683

ndeligia@etrovub.be

more info

Supervisor

Ir. Boris Joukovsky

+32 (0)2 629 2930

bjoukovs@etrovub.be

more info

- Contact person

- IRIS

- AVSP

- LAMI

- Contact person

- Thesis proposals

- ETRO Courses

- Contact person

- Spin-offs

- Know How

- Journals

- Conferences

- Books

- Vacancies

- News

- Events

- Press

Contact

ETRO Department

Tel: +32 2 629 29 30

©2022 • Vrije Universiteit Brussel • ETRO Dept. • Pleinlaan 2 • 1050 Brussels • Tel: +32 2 629 2930 (secretariat) • Fax: +32 2 629 2883 • WebmasterDisclaimer