CVPR Daily - Tuesday

Fabien Baradel is a PhD student at INSA Lyon in France. He speaks to us about his poster presentation today. The co- author of the paper are Christian Wolf, an associate professor at INSA-Lyon and LIRS/CITI, Julien Mille, associate professor at INSA Centre Val de Loire and LI Tours, and Graham Taylor, associate professor at University of Guelph. The research project is funded by the ANR/NSERC DeepVision. Fabien says his work is about human activity recognition , having a model which is able to infer what is going on in a video using only the RGB data. Its real-world application could be in video surveillance , or in human-robot interaction , to understand what is going on just in front of the robot e.g. what is the person doing in real time? In terms of human activity understanding, he says the real problem right now is that they are dealing with fine-grained understanding . There are some really local parts which are important and discriminative for the task, and right now, all the video understanding problems are more about understanding what’s going on in the context. This work is more about understanding a local part of the video. “…real-world applications could be in video surveillance, or in human-robot interaction, to understand what is going on just in front of the robot e.g. what is the person doing in real time? ” 14 Fabien Baradel Tuesday

RkJQdWJsaXNoZXIy NTc3NzU=