S-UAV Dataset
A Synthetic UAV (S-UAV) dataset for object detection and tracking that simulate real-world UAV videos, including variations in optical flow, occlusion, and segmentation.
Can YILMAZ - Bahri MARAŞ - Nafiz ARICA - Ayşın Baytan ERTÜZÜN
Embedded Files
The S-UAV dataset offers researchers a resource that will help them to train and test their own deep convolutional neural networks for object detection, tracking and classification in low altitude UAV videos. The dataset includes ground truths of optical flow, occlusion and segmentation for vehicles which have been monitored by UAV point of view.
OrnekSenaryolar.mp4Test Results of SwinT & YoloV7
Test Results of SwinT & YoloV7
mergedVideos.aviData Statistics
Data Statistics
The S-UAV dataset contains 16,285 video frames and 49,760 instances that have been automatically annotated using Unity 3D engine. The dataset consists of five object categories, including bus, car, lorry, truck, and van. Additionally, a smaller subversion of the 1920x1080 dataset has been created with a size of 250x250, which includes 99,520 cropped video frames and 49,760 instances.
250x250 Sized Dataset Outputs
Instance distribution according to weather conditions:
Folder Structure
Folder Structure
-1920x1080 Dataset
<Scenario Name x>
Original_images
<Tn>.png ...
Images
<Tn>.png ...
Masks
<Tn>.png ...
Flo Results
<Tn+1>.png, ...
<Tn+1>.flo, ...
Occlusion Results
<Tn+1>.png ...
Segmentation Results
<Tn>.png ...
Triangles.json
annotation.json
-250x250 Dataset
car_images
<Scenario Name x>
frame_<Tn>_<vehicle Id>_<vehicle count>_02.png,
frame_<Tn+1>_<vehicle Id>_<vehicle count>_01.png,
...
car_masks
<Scenario Name x>
frame_<Tn>_<vehicle Id>_<vehicle count>_02.png,
frame_<Tn+1>_<vehicle Id>_<vehicle count>_01.png,
...
coco
<Scenario Name x>
annotation.json
flow_f
<Scenario Name x>
frame_<Tn+1>_<vehicle Id>_<vehicle count>.flo, ...
flow_f_visual
<Scenario Name x>
frame_<Tn+1>_<vehicle Id>_<vehicle count>.png, ...
flow_b
<Scenario Name x>
frame_<Tn>_<vehicle Id>_<vehicle count>.flo, ...
flow_b_visual
<Scenario Name x>
frame_<Tn>_<vehicle Id>_<vehicle count>.png, ...
occ_f
<Scenario Name x>
frame_<Tn+1>_<vehicle Id>_<vehicle count>.png, ...
occ_b
<Scenario Name x>
frame_<Tn>_<vehicle Id>_<vehicle count>.png, ...
segmentation_results
<Scenario Name x>
frame_<Tn>_<vehicle Id>_<vehicle count>_02.png
frame_<Tn>_<vehicle Id>_<vehicle count>_01.png
...
* "x" represents the scenario folder -> {1, 2, 3, ...., 100}, "Tn" represents the frame ID -> {0, 1, 2, ..... M-1}, where M is the maximum frame ID of the x scenario. "Tn+1" represents the next frame after "Tn" -> {1, 2, ..... M}, where M is the maximum frame ID of the x scenario.
Download
Download
Cite - Bibtex
Cite - Bibtex
@INPROCEEDINGS{10223911, author={YILMAZ, Can and MARAŞ, Bahri and ARICA, Nafiz and ERTÜZÜN, Ayşın Baytan}, booktitle={2023 31st Signal Processing and Communications Applications Conference (SIU)}, title={Creation of Annotated Synthetic UAV Video Dataset for Object Detection and Tracking}, year={2023}, pages={1-4}, doi={10.1109/SIU59756.2023.10223911}}
@INPROCEEDINGS{10223740, author={YILMAZ, Can and MARAŞ, Bahri and YILMAZ, Görkem and CEYLAN, Göksu and HAMAMCIOĞLU, Önder and ARICA, Nafiz and ERTÜZÜN, Ayşın Baytan}, booktitle={2023 31st Signal Processing and Communications Applications Conference (SIU)}, title={Enhancing Object Detection Algorithms by Synthetic Aerial Images}, year={2023}, pages={1-4}, doi={10.1109/SIU59756.2023.10223740}}
@INPROCEEDINGS{10223740, author={YILMAZ, Can and MARAŞ, Bahri and YILMAZ, Görkem and CEYLAN, Göksu and HAMAMCIOĞLU, Önder and ARICA, Nafiz and ERTÜZÜN, Ayşın Baytan}, booktitle={2023 31st Signal Processing and Communications Applications Conference (SIU)}, title={Enhancing Object Detection Algorithms by Synthetic Aerial Images}, year={2023}, pages={1-4}, doi={10.1109/SIU59756.2023.10223740}}
Funding
Funding
TUBITAK Ardeb 1001
Page updated
Google Sites
Report abuse