Computer Vision News - December 2021

12 NeurIPS Workshop Preview strong and robust vision model using ImageNet. “ We’re very familiar with the ImageNet dataset and have been discussing it amongst ourselves for a while now and thinking about its future, ” he reveals. “ We thought, why don’t we organize our own event to get everyone together to talk about it? So many people work on or around ImageNet, but they have no place to meet and discuss ImageNet itself. This workshop will be a great opportunity for all of us to do that! ” The workshop will be a chance to reflect, regroup, and look to the future. Did we solve ImageNet? What have we learnt from it? What are the remaining challenges? What should the next generation of ImageNet-like benchmarks look like? Based on these questions, Sangdoo and his team have lined up a diverse range of invited speakers, interesting posters, and spotlight presentations. He hopes participants will gain a new perspective. IMAGENET: PAST, PRESENT, AND FUTURE Sangdoo Yun is a research scientist at Naver AI Lab in Korea and is a co- organizer of the first “ImageNet: past, present, and future” workshop at the upcoming NeurIPS conference. He speaks to us about what we can expect from the event. For more than a decade, ImageNet has played a crucial role in advances in computer vision, deep learning, and artificial intelligence. Originally created to train image classifiers, over the years it has become the go-to benchmark for modern architectures and training techniques . It has been used to maximum advantage for tasks such as few- shot learning, self-supervised learning, semi- supervised learning, and many more. Sangdoo and his team at Naver work on fundamental computer vision and machine learning and have been focusing on training a