Pedestrian detection at night is an under-represented yet important problem, where state-of-the-art detection algoritms fail, even when specifically trained with night-time data.
Following the success of the NightOwls Pedestrian Detection Challenge, ICCV 2019, we organize NightOwls Detetection Challenge 2020 as part of the Scalability in Autonomous Driving Workshop at CVPR 2020.
The NightOwls dataset and has 3 subsets - training, validation and testing. The training and validation subsets (images + annotations) are available below, the testing subset (images only) will be published at the beginning of the Testing phase of the competition. The number of submissions in the Testing phase is limited to 3.
The competition runs in three tracks.
The winners will be announced at the Scalability in Autonomous Driving Workshop, CVPR 2020.
Methods ranking is based on the standard Average Miss Rate metric used in the pedestrian detection literature [1], considering only targets within the Reasonable [1] set up (i.e. non-occluded targets with height >= 50px). The winning entry will be the method with the lowest Average Miss Rate. For Pedestrian - Single Frame and Pedestrian - Multi Frame tracks, we simply take the Average Miss Rate for the pedestrian class. For All Objects track, we average the Miss Rate over all classes, giving each object class an equal weight.
The server runs the same evaluation code from our NightOwls SDK, the evaluation can therefore easily also be run locally.
1. P. Dollár, C. Wojek, B. Schiele and P. Perona, Pedestrian Detection: A Benchmark, CVPR 2009, Miami, Florida
This dataset is made freely available to academic and non-academic entities for non-commercial purposes such as academic research, teaching, scientific publications, or personal experimentation. Permission is granted to use the data given that you agree:
Start: Jan. 1, 2020, midnight
Start: May 25, 2020, midnight
June 9, 2020, 11:59 p.m.
You must be logged in to participate in competitions.
Sign In