This repository provides annotations for the Aircraft Context Dataset, which focuses on training and evaluating classification, detection and segmentation in aerial domains. A thorough description can be found in the corresponding paper and supplementary material. The image data corresponding to the annotations must be downloaded and extracted from the original sources using the provided scripts. The required link list can be requested by email (please provide your full name, affiliation and intended use of the dataset).
After receiving the link list described above, it can be used as an input to the script setup_dataset.py to automatically download the videos and extract all annotated frames for either subset. Additionally, visualize_annotations.py is provided to visualize bounding boxes and semantic masks for all data samples.
All annotations are provided in the following format:
- bboxes: one csv file per frame with object instances defined as [left, top, right, bottom, objectId]
- params: per-image meta annotations for parameters described in section 3.2 of the paper [airborne, atmosphere, context1, context2, degredation, lighting, occluded, overexposure, underexposure, size, weight, [task1, task2, task3,] domain, propulsion, model].
- labelIds: semantic masks
The label specifications for classification, detection and segmentation are defined in datasets.py.
This subset currently consists of roughly 11k annotated frames and 2k semantic masks.
This subset will be released in Dezember 2021.
The Aircraft Context Dataset is released to academic and non-academic entities for non-commercial purposes such as academic research, teaching, scientific publications or personal experimentation (LICENCE).
If you use the dataset for your research, please use the following BibTeX entry:
@InProceedings{Steininger2021AircraftContextDataset,
author = {Steininger, Daniel and Widhalm, Verena and Simon, Julia and Kriegler, Andreas and Sulzbachner, Christoph},
title = {The Aircraft Context Dataset: Understanding and Optimizing Data Variability in Aerial Domains},
booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops},
month = {October},
year = {2021},
pages = {3823-3832}
}