***EVENT POSTPONED*** AlgoIL #11 - Computer Vision (non-deep) Algorithms!


Due to the coronavirus outbreak, the event is postponed.
Stay safe and healthy and we will have this event once everything is back to normal.

Do you think all computer vision problem are solved using deep learning? Well, think again and come for two great talks about algorithms in the field of computer vision. We will hear how weighted random sampling algorithms improve panorama stitching and about visual localization of drones.

Thanks Trax for hosting this event!

18:00 Gathering and snacks
18:30 Opening words
18:40 Maria Kushnir - Weighted random sampling algorithms - from theory to application
19:25 Short break
19:35 Nomi Vinokurov - Where is my drone? Accurate 6DOF localization
20:20 Time to mingle a little more

Both talks will be held in Hebrew.
The talks will be filmed and uploaded to our YouTube channel: http://bit.ly/AlgoILYoutube

Maria Kushnir - Weighted random sampling algorithms - from theory to application

In order to take virtual walks on any street in the world, explore historical landmarks or find the hottest shops, restaurants, and hotels, you'd probably use the Google Street View feature on Google Maps. This is a feature that provides panoramic views of various streets in the world, and is created from several image sources from different viewpoints, merged together.

At Trax, we digitize retail shelves by using computer vision, in order to understand the exact product arrangement on the store's shelves and unlock usages of such data. This involves capturing numerous images across hundreds of aisles, each containing many products. One of the most compelling means of visualizing data of this form is via multi-perspective panoramas. In simple terms, this process amounts to extracting individual images taken from different viewpoints and devices and then aligning them to generate the single store view as a final panorama.

In this talk, I will give a short introduction to this process of mosaicing - or stitching, and will then present the challenges and their computational complexity in the retail domain, where identical products are often placed next to each other. Finally, I'll describe our tailor-made variations of the popular approaches, enabled by improvements in sampling strategies.

About the speaker:
Maria Kushnir received her Ph.D. degree in computer vision from Haifa University and M.Sc. degree in Theoretical Physics from the Technion. In 2016 she joined Trax (traxretail.com) as an algorithm developer. She is currently working on multiple view geometry and machine learning applications, that convert store images into shelf insights.

Nomi Vinokurov - Where is my drone? Accurate 6DOF localization

ClearVuze was an autonomous cinematography company. We were developing AI solutions to fully automate filming the world – from hiking to sports, to beach days and pool parties. The idea was to enable the consumers to get the maximum from their drone.

We do that by moving an AI-powered drone into the right place at the right instant to capture the best shots. Generating those best shots requires precise localization of the drone camera, its location and rotation.

Join me for this talk where we will discuss the many challenges of drone visual localization, and offer some solutions.

About Nomi Vinokurov:
While studying Psychobiology BA, I discovered the huge power computers have, and in particular the machine learning field. I dove into the machine learning field, and made my Computer Science MS.c thesis in the fields of Novelty Detection, Machine Learning and Computer Vision. Since then I have been working on various computer vision problems, as 3D semantic segmentation, stabling videos generated by drone, and drone localization.