Computer Vision Track, Winter School 2017, Computer Science @ UCU

Given by


kulatelogo.jpg cmp80.jpg

Course 1: Particular-instance Retrieval

This course gives the theory and practical exercises needed to build a state-of-the-art bag-of-words image-retrieval system. Specifically, the course focuses on the methods necessary for particular-instance retrieval, which is the task of ranking the most visually similar instances of a query object (e.g., art work, landmark, or logo) from a very large-scale image database, typically tens-of-millions of images or more.


James Pritts ( from Czech Technical University in Prague


Date Topic Practical Slides & videos
15.1 MATLAB preliminaries Lab 1
16.1 Introduction to Image and particular-instance retrieval Retreival intro, Retreival review, Detail Mining
16.1 Local Features Intro, Image Processing, Gaussian Derivatives Lab 2 Local features I, Gaussian derivatives
16.1 Local Feature Extraction, Scale Space Lab 3.1 Local features II
16.1 Local Feature Description for Image Retrieval and Wide-baseline matching Lab 3.2 Local features III
17.1 RANSAC and Two-view projective geometry Lab 3.3 RANSAC
17.1 Cluster analysis, K-means method, vector quantization Lab 4 K-Means
18.1 Bag-of-words, TF-IDF, Inverted file Lab 5 Min-hash I , Min-hash II
18.1 Min-hash, Spatial Verification, Query Expansion

Please, don`t delete submitted homework after end of this course. You will need you retrieval homework ( Labs 3.* ) for tracking course assignments

Course 2: Tracking

Visual object tracking is one of the main problems in computer vision. It has practical applications in: autonomous driving, human-computer interaction, film post production, image stabilization, measurements (e.g, sports), etc. I am going to give an overview of the state-of-art tracking algorithms, then focus on their practical implementation.


Dmytro Mishkin ( from Czech Technical University in Prague


Date Topic Practical Slides
20.1 What is tracking. Optical flow. KLT tracker Lab 8 Day1
20.1 Tracking by correspondence Lab 9
20.1 Mean-Shift tracker.
21.1 Convolutional networks recap. CNN Design choices. Fine-tuning of CNNs. Day2
21.1 CNN for tracking. MDNet tracker. Siamese FCN tracker Lab 10
22.1 Correlation filters trackers family. Lab 11 Day3
22.1 Online discriminative tracking. Long-term trackers.
22.1 How to evaluate trackers.

Students have two choices for practical part of the course. First - is to complete all Labs. Second - to complete 3-day project from the following list (additional links TBD):

  1. Develop eye tracker. Select eye detection or tracking algorithm and go on. You can use any code you download.
  2. Develop “hand writing in the air” via tracking algorithm. You can use any code you download.
  3. Develop football tactics analyser. Start from estimating players and ball trajectories during the match.You can use any code you download.
  4. Lets augment our reality. Develop an app to add cup content name to the cup, as if it were painted on cup. You can use any code you download.
  5. Implement STAPLE tracker from scratch. You cannot use 3rd party implementations
  6. Implement MDNet tracker from scratch. You cannot use 3rd party implementations
  7. Implement Continuous Convolution Operator Tracker from scratch. You cannot use 3rd party implementations
  8. Implement any tracking paper you like from timeframe (2013 - … now). Submit results to the VOT. You cannot use 3rd party implementations

Basic info

The labs require you to implement some selected core components for two classic and open problems in computer vision: particular-instance retrieval and tracking. Your implementations will be tested with different critical tasks from these two topics. Each day a new set of assignments will be introduced, and you are expected to complete the implementations before the end of the day. The lecture sessions will introduce the theory that justifies the methods used to complete the practical tasks given by the labs. The lab sections are devoted to individual interactions between teams of students and the teacher.


Basic knowledge of the folowing: MATLAB, please refer to Lab 1, linear algebra, statistics, vector calculus, classic data structures and algorithms (think Corman et al), and some familiarity and experience with machine learning. However, the labs are designed to be as self-contained as possible as time allows.

  • Please attempt the first lab, Lab 1, before the start of the course.
courses/ucuws17/start.txt · Last modified: 2017/01/27 22:51 by mishkdmy