CADOT Dataset
CADOT images are collected from the
National Institute of Geographic and Forest Information (IGN) . All images are stored in .jpg format. The dataset includes:
- 4,628 high-resolution (500×500 pixels) aerial images from the Paris region.
- 106,691 object annotations across 14 categories, including vehicles, buildings, playgrounds, and more.
- Complex urban scenes with occlusions and varying object sizes, offering a challenging benchmark for object detection models.
Object Categories
CADOT contains 14 annotated object categories:
Full court, playing area & boundaries. Excludes bleachers.
Roof outline including overhangs. No adjacent trees.
Only marked lines. No sidewalks or roads.
Complete playing area. Excludes audience zones.
Tombstone rows only. Excludes vegetation or lanes.
Includes trailers/mounts. No overlaps.
Full vehicle body. Avoids shadows.
Includes equipment area. Excludes benches or paths.
Encloses circle & island, not vehicle lanes.
Ship only. Excludes water, piers, docks.
Car body with wheels. No shadows.
Only water surface. No decks or boards.
Playable areas and lines. Excludes fencing.
Carriages or full train along track.
Annotation Format
We use the COCO format annotation for our CADOT dataset, ensuring compatibility with popular object detection frameworks. This format facilitates seamless integration into deep learning pipelines by organizing data into categories, images, and object-level annotations. CADOT is structured to support object detection, classification, and other vision tasks.
Dataset Structure
The CADOT dataset is split into train,
validation, and test sets.
The annotation files are provided for training and validation.
📦 Download here:
CADOT_Dataset.zip
