Winning the Robust Vision Challenge 2020

We are proud to announce that BrainLinks-BrainTools researchers Prof. Abhinav Valada and Rohit Mohan have won the Robust Vision Challenge 2020 - Panoptic Segmentation, that was held in conjunction with the European Conference on Computer Vision (ECCV), the most important computer vision conference in Europe. The challenge was sponsored by industry leaders such as Apple, Google, Intel, Naver Labs, and Mapillary.

The purpose of this challenge is to foster the development of vision systems that are robust and consequently perform well on a variety of datasets with different characteristics, e.g., indoor vs. outdoor, real vs. synthetic, sunny vs.bad weather, different sensors. The challenge required participants to train a single model over a joint dataset comprising of six different benchmarks (COCO, Cityscapes, Mapillary Vistas, VIPER, WildDash, KITTI).

Panoptic segmentation is task that enables the simultaneous recognition of distinct foreground objects such as people, cyclists, or cars, while also labeling pixels in the image background with classes such as road, sky, or grass. Prof. Valada and his team recently developed the EfficientPS model that broke multiple records in this task. For the Robust Vision Challenge 2020, the BrainLinks-BrainTools researchers developed a lightweight version of their EfficientPS model that achieved the first place in this challenge. Read the winning report here.