ACRV Robotic Vision Challenge 1 Validation Data
This is the validation data for the first Australian Centre for Robotic Vision (ACRV) Robotic Vision Challenge. It consists of 4 rendered video sequences in which participants must detect objects, with both spatial and semantic uncertainty. This validation data also contains pixel-level ground-truth annotations for the 30 classes of object to be detected.
For more details, see https://competitions.codalab.org/competitions/21727
The sequence contains synthetic data generated using Unreal Engine 4. For more details on its generation, see the workshop paper: https://arxiv.org/abs/1903.07840.
Data files consist of a single zip file, containing the frames and ground-truth data for 4 video sequences. The frames folder within the zip file contains a separate folder of .png images for each of the 4 video sequences. The ground_truth folder within the zip file contains a seperate folder containing mask .png images and a single labels.json file for each video sequence.
For information on how to read the ground-truth format and use it for evaluation, see https://github.com/jskinn/rvchallenge-evaluation.
Contact: contact@roboticvisionchallenge.org.
