1 minute read time.


The Vision & Imaging Network found itself rubbing shoulders with tech giants Microsoft Research. Toshiba and Snapchat at the recent 2017 British Machine Vision Conference aka BMVC 17!


The impressive conference, hosted by Imperial College, London, ran over four days, with most of the conference sessions and exhibitors based at the impressive Royal Geographical Society premises in leafy Kensington, just a stone’s throw from Hyde Park.

Over the course of four days the conference delivered an impressive range of over 40 oral presentations, 140 posters, keynotes, tutorials and workshops. Phew!

Keynotes were delivered by Facebook’s Richard Szeliski, on Visual Reconstruction and Image-Based Rendering; and Pietro Perona, California Institute of Technology, on Visipedia – A Universal Visual Expert.


At the conference Vision & Imaging Network Committee member Prof. Majid Mirmehdi, Uni. of Bristol joined me in reviewing the posters; he felt that the following two posters really stood out.


Oxford University’s Arsha Nagrani presented From Benedict Cumberbatch to Sherlock Holmes: Character Identification in TV series without a script which looks at the identification of on-screen characters without the need of using a script. The proposed technology uses both vision and a voice modality to identify characters, and tests have shown that the technology is successful even when applied to dark scenes. The technology was also successfully applied to Casablanca, filmed in 1942 and considered ‘poor quality’ by today’s standards.

Nagrani envisages that the technology could be used in various commercial scenarios, such as intelligent fast-forwarding, news stories and fact checking; and Amazon X-ray.


Another highlighted paper was Scott Wehrwein (Cornell University) and Richard Szeliski’s poster on Video Segmentation with Background Motion Models. The broadcast industry has been increasingly interested in being able to edit specific objects within a scene and the poster addressed the ability to pull out specific objects from video backgrounds using RANSAC.

 

Object based editing ha been gaining increasing interest in the world of broadcast and with IBC 17 currently taking place in Amsterdam it’ll be an interesting technology to watch develop.