Below is a brief description of our recent externally funded research projects. All our projects result to scientific publications, BSc/MSc/PhD theses and technical reports, and therefore more information such as code snippets, datasets and videos can be found from the publications page (each publication item contains links to all available material).

More project related information (Github and Bitbucket code repositories, datasets) can be found from the publications page.

Machine Learning for Robot Control

The main goal of this project is to develop novel Learning by Demonstration methods for heavy machinery. The fundamental idea is that a human expert provides examples by operating the heavy machine and then these “trajectories” are learnt by machine learning techniques. The main result of this project is a novel neural network controller that learns from real data and operates then autonomously.

Funding: Academy of Finland (Profiling – Profi1)


A Lot of Classes

This project investigates novel approaches to learn and detect a large number of visual classes. The main goal is to develop methods that are able to effectively learn visual object detectors from a small number of examples and to efficiently search these classes from unseen images. Visual object detection is one of the fundamental problems and grand challenges of computer vision. One of the main results of this project is a novel computing paradigm – Visual Similarity Graph – that can represent multiple semantic connections simultaneously.

Funding: Academy of Finland

Large Scale 3D

This project investigates computer vision technologies that combine street view snapshots and 3D point clouds to 1) provide augmented reality experience to mobile phone users and 2) provide accurate pose estimation of a device or a car in urban environments. One of the main results of the project is a large scale comparison of 3D  camera pose estimation methods and a novel hybrid pose estimation algorithm that combines a feature-based and a mutual-information based method and achieves state-of-the-art.

Funding: Nokia Technologies, Graduate School