The 2nd CVPR DataCV Challenge
Overview
The 2nd DataCV Challenge is held in conjunction with the CVPR 2024 Visual Dataset Understanding workshop. It is the first of its kind in the community, where we focus on searching training sets for various targets. The competition is held via CodaLab and includes two phases.
Important Dates and News
Monday, February 5th, 2024: Training and validation data release
Monday, February 12th, 2024: Release of the evaluation server
Saturday, March 9th, 2024: Test data release
Saturday, March 16th, 2024: Result submission closed
Midnight 23:59 March 20th, 2024 (Pacific Time): Workshop paper submission closed
Midnight 23:59 March 27th, 2024 (Pacific Time): Supplementary material submission closed
Thursday, April 11th, 2024: Author notification of paper acceptance
Saturday, April 13th, 2024: Camera ready.
Task Overview
Training data search in object detection. In the training set search challenge, we aim to search small-scale, yet highly effective training sets from a large-scale data pool such that a competitive target-specific model can be obtained.
Unlike traditional challenges where the training set is fixed, allowing participants to optimize their model or algorithm, in our 'training set search' challenge, we keep the model or algorithm fixed and allow participants to optimize the training set.
Datasets
Overview: The datasets for a training set search task consist of two components: the data source pool and the target dataset. The data source pool will be formed by combining multiple existing detection datasets, such as MSCOCO. For the target dataset, we will newly introduce a database consisting of a diverse range of detection environments for the development and evaluation of scene adaptive object detection. Specifically, this dataset will be recorded across 100 countries, capturing scenes at different times of the day and throughout the year. This dataset exhibits high diversity and poses significant challenges to existing detection systems.
Availability: The source pool, target training set and test set A are now publicly available on Github. The test set B remains reserved for the determination of challenge awards. We offer an evaluation server for calculating test accuracy, and will be released soon.
Evaluation website: We use CodaLab as our competition platform. The evaluation has been released!
Ethical considerations: We blurred the human face and vehicle license plate before releasing the datasets, to ensure individual privacy. Furthermore, we follow the practice of (Asano et al., NeurIPS 2021) to validate the copyright before distributing the datasets.
Organizers
Australian National University
Northeastern University
Australian National University
Narrabundah College