The Computer Vision Center organised this year three challenges in the framework of the workshop series on Perception Beyond the Visible Spectrum
The Computer Vision and Pattern Recognition conference (CVPR) is the world’s biggest event on Computer Vision Science. The event, which took place this year from June 19 to June 24, comprised several co-located workshops and courses besides the main conference. In fact, there were over a hundred workshops organized in conjunction with CVPR. There is one, though, the workshop series on Perception Beyond the Visible Spectrum (IEEE PBVS), that since its inception in 2004 has been key for the CVPR community and that this year celebrated its eighteenth edition.
In this year's edition, Dr Angel D. Sappa, Senior Researcher of the Computer Vision Center (CVC), as Program Co-Chair and Challenge Chair within the Organising Committee, has coordinated three challenges in the framework of the IEEE PBVS'22 workshop. All relevant information about the challenges can be found at CodaLab, which is the platform that supported the competition.
The 3rd Thermal Image Super-Resolution Challenge aims to introduce state-of-the-art approaches for the thermal image super-resolution problem, as well as, to promote a novel thermal image dataset to be used as a benchmark by the community. Almost a hundred teams competed in this challenge that consisted in obtaining super-resolution images from a given thermal image dataset. This shows that the topic is of interest to the community and it is expected to encourage future research in this area.
A paper presenting the results of the challenge has been published Open Access. Read the publication here:
The Semi-Supervised Hyperspectral Object Detection Challenge provides participants with images which have only 10% of their data labelled. Participants are expected to develop frameworks that can solve the semi-supervised object detection problem. Thirty-eight individuals registered to this challenge on its first edition.
A paper summarising the top contributions has been published Open Access. Read the publication here:
And, the Multi-modal Aerial View Object Classification Challenge consists in predicting the class label of an aerial low-resolution image. It aims to inspire research that develops methods for building recognition models that use two types of imagery: synthetic aperture radar (SAR) and electro-optical (EO) imagery. The challenge was divided in two tracks (track1 & track2) and had a high participation rate with a total of 159 teams taking part in the competition.
A paper discussing the top performing methods has been published Open Access. Read the publication here:
Out of all three challenges, MAVOC counted, among other prize categories, with a cash prize for the first, second, and third place for each of its respective tracks. The prize was possible thanks to the support and courtesy of the Wright Brothers Institute, known to be the United States Air Force Research Laboratory, and the Computer Vision Center was the institution in charge of handling the award to the winners. Check the complete winner list for each challenge at the workshop's website.
IEEE PBVS, in conjunction with CVPR, is a leading meeting that provides exceptional value for researchers, academics, and students worldwide. Join next year's edition from June 17 to June 23 in Vancouver, Canada.