tulips

ICCV - Day 7

Back to workshops today. One of those “forward-looking” days where there’s not much that is related to my current research, but lots to take in for future research!

I skipped around between two workshops: one on “‘socially intelligent surveillance and monitoring, which interests me quite a bit for future work, and another on live dense reconstruction from moving cameras (note: live), which is by far one of the most popular workshops of the day (with a room full of cameras, Kinect sensors and quad-core machines!) Here’s a picture of Shahram Izadi of Microsoft Research demonstrating a live scene reconstruction using a Kinect sensor.

Check out his website for more mind-blowing touch-based technology from Microsoft Research!

I have spent quite awhile working on my current topic of spatio-temporal representations for video-based face recognition, and as I said before, I have probably lost track of what are the current trending areas in computer vision. A big portion of my personal effort goes into getting back in touch with the computer vision community (which is one of those great benefits of attending top conferences, I believe), and getting ideas of what projects I could potentially work on during post-PhD. (It’s of common knowledge that whatever these people do today will probably only take flight in Malaysia a few years later…)

But this year’s ICCV has seen the field took giant steps towards 1) activity-related understanding (no longer confined to gestures alone) including recognition, and 2) large-scale multi-object classification using both features and attributes. Related periphery works, such as object segmentation, spatio-temporal representations (for videos in particular), topical models and descriptors are still of immense interest as they contribute greatly towards the success of these higher level works.