Panoptic segmentation task combines semantic and instance segmentation. The goal of panoptic segmentation is to map each pixel of an image to a pair (semantic class, instance id). This challenge is formulated as follows. Given an input sequence of RGB/RGB-D frames, predict the panoptic segmentation labelling for each frame. One can utilize previously seen frames, e.g. create and use an internal scene representation of the house.

Evaluation protocol

We follow the evaluation protocol of COCO Panoptic Segmentation Challenge.

Result format

We use the same data format as in panoptic segmentation part of COCO dataset.

Dataset will be available after July 15 2019

Dev toolkit will be available after July 15 2019