Detail of the student project

List
Topic:BENCHMARK WEBOVÁ STRÁNKA pro vyhodnocení algoritmů počítačového vidění na syntetické CGI datové sadě
Department:Katedra kybernetiky
Supervisor:Mgr. Jan Šochman, Ph.D.
Announce as:Diplomová práce, Bakalářská práce, Semestrální projekt
Description:Background: Together with the computer graphics group, we are developing a synthetic CGI driving dataset with realistic rendering. Apart from the images themselves, it also contains additional  modalities like depth in every pixel, object locations, motion fields, ... This data is meant for training and evaluation of CNN-based computer vision algorithms.

The goal of the project is to develop:

1. A benchmark web site for evaluation of different computer vision algorithms on the test part of this dataset. The site itself will be derived from an existing benchmark [1]. This requires knowledge of Python/Django [2].

2. Several relevant metrics useful for problems like optical flow estimation, dense object detection, motion segmentation, ... An inspiration will be drawn from sites [3], [4] or [5].

3. Python scripts for algorithm developers, simplifying their task of uploading their results.

[1] https://bop.felk.cvut.cz
[2] https://www.djangoproject.com/
[3] http://www.cvlibs.net/datasets/kitti/index.php
[4] http://sintel.is.tue.mpg.de/
[5] http://www.robustvision.net/index.php
Responsible person: Petr Pošík