Invited Speakers

Invited speakers

  • Cordelia SchmidINRIA
    • Title: Recent Progress in Spatio-Temporal Action Location
  • Abhinav GuptaCMU
    • TitleSelf-supervised Learning of Visual Representations
    • AbstractIn this talk, I will discuss our effort in self-supervised learning of visual representations (ConvNets). First, I will talk about how we can use context (relative spatial location) as a supervision to train ConvNets. Next, I will talk about how to extend how the idea of self-supervised learning to videos. Specifically,  we will show how to use you-tube videos to learn ConvNet representations by learning object instance invariances. Finally, I will talk about how we can learn visual representation in presence of noisy-web labels. I will also briefly mention our efforts in using minimal human supervision to improve the multi-label classification and the newly introduced Open-Image dataset.
  • Raquel UrtasunUniversity of Toronto
    • TitleTorontoCity Benchmark: Towards Building Large Scale 3D Models of the World
  • Zehan Wang, Twitter
    • Title: Scaling-up: Image Super-resolution and Compression for the masses.
    • Abstract: Rich media consumption is growing rapidly and we are getting ever more accustomed to high quality visual experiences (e.g HD, 4K, VR). Yet at the same time, the capacity to deliver these experiences has not grown nearly as fast - we've all experienced the annoyance of slow video buffering, and low visual quality when streaming on poor connections / mobile networks. Are there ways in which computer vision and machine learning methods can better alleviate the strains outside of traditional media codecs? We introduce some new ideas and recent work in this space which are centred around enabling better visual experiences whilst reducing bandwidth requirements.
  • Antonio Torralba, MIT
    • Title: Learning to see objects by listening