| --- |
| language: |
| - en |
| license: cc-by-nc-4.0 |
| task_categories: |
| - object-detection |
| pretty_name: MovingDroneCrowd++ |
| tags: |
| - crowd-counting |
| - tracking |
| - drone-imagery |
| --- |
| |
| # MovingDroneCrowd++ |
|
|
| MovingDroneCrowd++ is a large-scale video-level dataset dedicated to dense crowd counting and tracking from fast-moving drones. It was introduced in the paper [Video Individual Counting and Tracking from Moving Drones: A Benchmark and Methods](https://huggingface.co/papers/2601.12500) (Extended version) and [Video Individual Counting for Moving Drones](https://huggingface.co/papers/2503.10701) (ICCV 2025). |
|
|
| - **GitHub:** [fyw1999/MovingDroneCrowd](https://github.com/fyw1999/MovingDroneCrowd) |
| - **Paper (Extended):** [Video Individual Counting and Tracking from Moving Drones: A Benchmark and Methods](https://huggingface.co/papers/2601.12500) |
| - **Paper (ICCV 2025):** [Video Individual Counting for Moving Drones](https://huggingface.co/papers/2503.10701) |
|
|
| ## Dataset Description |
|
|
| MovingDroneCrowd++ captures dense crowds under diverse flight altitudes, camera angles, and illumination conditions. Each pedestrian head is annotated with a bounding box and an identity ID across frames, enabling tasks such as: |
| - Video individual counting (VIC) |
| - Inflow/outflow analysis |
| - Dense crowd tracking |
|
|
| ## Dataset Structure |
|
|
| The expected directory structure for the dataset is as follows: |
|
|
| ```text |
| MovingDroneCrowd++/ |
| +-- frames/ |
| | +-- scene_1/ |
| | +-- 1/ |
| | +-- 1.jpg |
| | +-- 2.jpg |
| | +-- ... |
| +-- annotations/ |
| | +-- scene_1/ |
| | +-- 1.csv |
| +-- val.txt |
| +-- train.txt |
| +-- test.txt |
| +-- scene_labels.txt |
| +-- MDC_val.txt |
| +-- MDC_train.txt |
| +-- MDC_test.txt |
| ``` |
|
|
| ## Annotation Format |
|
|
| Each annotation row follows the MOT-style format: |
| ```text |
| frame_id, person_id, x, y, w, h, -1, -1, -1, -1 |
| ``` |
|
|
| The first column is the frame index, the second is the pedestrian identity, and the third to sixth columns are the head bounding box `(x, y, w, h)`. |
|
|
| ## Sample Usage |
|
|
| You can download the dataset using the Hugging Face CLI: |
|
|
| ```bash |
| huggingface-cli download fyw1999/MovingDroneCrowd --repo-type dataset --local-dir /path/to/MovingDroneCrowd++ |
| ``` |
|
|
| ## Citation |
|
|
| ```bibtex |
| @article{MDC++_GD3A, |
| title={Video Individual Counting and Tracking from Moving Drones: A Benchmark and Methods}, |
| author={Fan, Yaowu and Wan, Jia and Han, Tao and Ma, Andy J. and Ouyang, Wanli and Chan, Antoni B.}, |
| journal={arXiv preprint arXiv:2601.12500}, |
| year={2026} |
| } |
| |
| @inproceedings{MDC_SDNet, |
| title={Video Individual Counting for Moving Drones}, |
| author={Fan, Yaowu and Wan, Jia bit and Han, Tao and Chan, Antoni B. and Ma, Andy J.}, |
| booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)}, |
| month={October}, |
| year={2025}, |
| pages={12284--12293} |
| } |
| ``` |
