Seminars

IMPACT Seminars

Please follow the seminar page on IMPACT Project website.

AAG Seminars

Please subscribe to the Google Calendar  or iCal.

Upcoming Seminars

2019-07-24, 11:00 – 12:00, B-633, Vladimir Petrik. Reinforcement learning and its applications

Abstract:
In the mini-seminar, I will briefly describe the reinforcement learning (RL) framework with the focus on model-based RL. The model-based RL builds an intermediate model from observations and trains the controller on it. This intermediate layer increases the data-efficiency but at the same time, it increases the computational cost. I will show the tasks on which I have applied the RL in the past and suggest future extensions. The seminar will be followed by a discussion of these future extensions/directions.

Past Seminars

2019-07-18, 11:00 – 12:30 Timothy Duff. Intro to homotopy continuation with a view towards minimal problems
2019-07-17, 14:00 – 15:30 Kathlen Kohn. Point-Line Minimal Problems in Complete Multi-View Visibility
2019-07-16, 11:00 – 12:30 Andrea Fussielo. Synchronisation: from pairwise measures to global values
2019-03-05, 14:00 – 15:30 Simon Telen. Stabilized algebraic methods for numerical root finding
2019-01-23, 11:00 – 12:00 Mircea Cimpoi. [Internal seminar] NeurIPS 2018 Summary (internal link) – and Neighborhood consensus networks
2018-10-05, 11:30 – 12:00 Mateusz Kozinski. Learning to Segment 3D Linear Structures Using Only 2D Annotations
2018-06-11, 10:00 – 12:00  Mircea Cimpoi. [Reading group] Learning to Find Good Correspondences
2018-06-05, 14:00 – 16:00  Stanislav Steidl. [Reading group] DSAC-Differentiable RANSAC for camera localization
2018-05-31, 10:00 – 12:00  Pierre-Yves Masse. A shortest path introduction to reinforcement learning
2018-05-24, 10:00 – 11:00 Jiri Sedlar. Estimating 3D person-object interactions with contacts
2018-05-24, 11:00 – 12:00  Georgy Ponimatkin. Possible applications of deep learning for ultrarelativistic heavy-ion collisions 
2018-05-21, 12:45 – 14:15 Michal Jancosek. Automatic reconstruction of 3D models @ capturingreality.com
2018-05-17, 10:00 – 12:00
Pavel Trutman. Solving polynomial equations by optimization
2018-05-10, 10:00 – 12:00 Tomas Pajdla. What are Groebner bases and how to solve systems of polynomial equations II
2018-05-04, 14:00 – 16:00 Tomas Pajdla. What are Groebner bases and how to solve systems of polynomial equations I
2018-04-23, 10:00 – 12:00 Mircea Cimpoi. Introduction to CNNs
2018-04-17, 14:00 – 16:00 James Pritts. Radially-Distorted Conjugate Translations
2018-04-10, 10:00 – 12:00 Federica Arrigoni. Consistent multi-view matching by spectral decomposition
2018-03-23, 10:00 – 12:00 Ludovic Magerand. From Incremental To Hierarchical Projective SfM

Abstracts

2019-03-05, 14:00 – 15:30 Simon Telen. Stabilized algebraic methods for numerical root finding
Abstract
We consider the problem of finding the isolated points defined by an ideal in a ring of (Laurent) polynomials with complex coefficients. Algebraic approaches for solving this use rewriting techniques modulo the ideal to reduce the problem to a univariate root finding or eigenvalue problem. We introduce a general framework for algebraic solvers in which it is possible to stabilize the computations in finite precision arithmetic. The framework is based on truncated normal forms (TNFs), which generalize Groebner and border bases. The stabilization comes from a ‘good’ choice of basis for the quotient algebra of the ideal and from compactification of the solution space.

2018-05-21, 12:45 – 14:15 Michal Jancosek Localization of the photography in the 3D space, automatic reconstruction of the accurate 3D model and state-of-the-art of modern photogrammetry (Karlovo namesti 3, Buiding E, 1st floor, room K-127 (https://www.youtube.com/watch?v=cwxHhPTC3KY)
Abstract
Extensive academic and commercial research, significant improvement of the quality and availability of cameras has brought its benefits in a form of robust methods that allow reconstruction of highly precise 3D models using easily accessible gadgets, as computer + standard camera/smartphone. In this session, the graduate of the CTU in Prague, RNDr. Michal Jančošek PhD., one of the co-founders of the Slovak company Capturing Reality, will introduce the current state-of-the-art in automatic extraction of 3D models using images, laser scans, UAV or synchronized camera rigs. Michal will talk about the changing possibilities and what new challenges has this opened up. He will also introduce their Software Development Kit (RealityCapture Engine) that allows smooth use of their in-house developed algorithms for automatic camera calibration, calculation of camera poses and orientation, 3D model reconstruction to the triangular mesh form, creating high quality per-vertex colors and texturing for the model and offers also algorithms for scene simplification, render and export. In brief, he will demonstrate the use of the Engine and will outline the number of possible research and application thesis that it could be applied on.