We are delighted to announce today an exciting line-up for our first 4x London Computer Vision Meetups - as follows:
The most important components when applying neural networks for image super resolution (SR) are: training data, network architecture, objective function. Deep residual networks currently provide the most accurate reconstructions in terms of peak signal-to-noise ratio. However, due to the commonly used, mean-square-error based loss functions they struggle to resolve high-frequency details. Reconstructions are often perceptually unsatisfying in the sense that they fail to match the fidelity expected at the higher resolution. By redefining the objective function based on generative adversarial networks (GANs) we achieved a step-change in the perceived quality of super resolved images.
Bio: Christian Ledig (@LedigChr) is a Computer Vision Researcher at Magic Pony, Twitter in London. He received a PhD from Imperial College London in 2015, where he was working on medical image analysis under the supervision of Prof. Daniel Rueckert. His current research focuses on deep learning approaches and generative models, in particular generative adversarial networks, for image and video super-resolution.
From learning to classify images to learning to play video games, end-to-end deep learning has become the dominant paradigm in computer vision. It has surpassed many hand engineered techniques which relied on principled geometry. However, these deep learning models are largely big black-boxes which we struggle to understand. They require ginormous quantities of data and struggle to generalize. Perhaps, they are often naive and missing a principled understanding? In this talk, I'm going to argue that many of the next advances in computer vision with deep learning will come from insights to geometry. By geometry, I refer to representations such as depth, volume, shape, pose, disparity, motion and optical flow. I will reflect on how we can draw insights from the decades of work into classical computer vision geometry. I will show how we can use this understanding to better design deep convolutional neural network architectures, loss functions and frameworks. Specifically, this talk will cover an overview of my work in semantic segmentation, camera relocalisation, stereo depth estimation and multi-task learning for scene understanding.
Bio: Alex Kendall co-founded and is CTO of Wayve.ai in addition to holding a Research Fellowship at Trinity College at the University of Cambridge in the United Kingdom. He graduated with a Bachelor of Engineering with First Class Honours in 2013 from the University of Auckland, New Zealand. In 2014, he was awarded a Woolf Fisher Scholarship to study towards a Ph.D. at the University of Cambridge where he was supervised by Prof. Roberto Cipolla.
Alex’s research investigates applications of deep learning for the perception and control of robots. He has developed computer vision algorithms to enable autonomous vehicles to understand complex and dynamic scenes. He has published multiple scientific papers in many of the top-tier computer vision, robotics and artificial intelligence conferences such as ICCV, CVPR, IJCAI, ICRA and NIPS. His technology has been used to control self-driving cars with Toyota Research Institute, power smart-city infrastructure with Vivacity and enable next-generation drone flight with Skydio.
Assuming all goes well, location for all 4 x Meetups will be:
One Alfred Place
Approx. timing for all Meetups will be:
18:00: Doors open
19:00 - 19:30: First talk
19:30 - 20:00: Break & networking
20:00 - 20:30: Second talk
23:00 - Close
A limited amount of pizza and beer will be provided on a ‘first-come-first-served’ basis - and food will be available to pre-order or order on the day (e.g. Club Sandwich; Burger; Chicken Burger; Chicken Caesar Salad - £12 each). One Alfred Place also has a full bar that will be open until 2300.
FYI - for the foreseeable future, our plan is to continue arranging a strong Meetup every 2nd month (continuing May, July, …) each with two speakers - ideally, one with more of an academic/research focus and one with more of a commercial / implementation focus. If you are interested in speaking at a future Meetup, please get in touch by emailing us at firstname.lastname@example.org.