IR Sensor
Instance
Segmentation
Dataset
2025
Dataset Overview
-
This dataset is designed for autonomous driving environments and was captured using the Hanwha Systems' QuantumRed Thermal Imaging Camera across diverse conditions, times, and locations.
-
The 2025 instance segmentation dataset is derived by sampling a portion of the dataset using in the 2024 Object Detection Challenge.
-
Classes
-
Person(0), Car(1), Truck(2), Bus(3), Bicycle(4) Bike(5), Extra_vehicle(6), Dog(7)
-
Subset | Quantity | Image Availablity | Label Availablity | Publication |
|---|---|---|---|---|
Train | 5,937 | O | O | O |
Validation | 600 | O | O | O |
Test open | 1,636 | O | X | O |
Test private | 1,629 | X | X | X |
Annotation Format
-
Annotation are structured in the MSCOCO format, with each dataset represented by a single JSON file.
{
"images":[
{
"file_name" : Image name(str),
"height" : Image height(int),
"width" : Image width(int),
"id" : Unique image ID(int)
}, ...
],
"annotations":[
{
"id" : Unique object ID(int),
"image_id": ID of the image containing the object(int),
"bbox" : [
X-coordinate of the top-left points of the object's bounding box(int),
Y-coordinate of the top-left points of the object's bounding box(int),
Width of the object's bounding box(int),
Height of the object's bounding box(int)
],
"area" : Area of the Object in Pixels (int),
"category_id" : Object class number(int),
"segmentations" : [
[
The X-coordinate of the points that make up the polygon,
The Y-coordinate of the points that make up the polygon,
The X-coordinate of the points that make up the polygon,
The Y-coordinate of the points that make up the polygon,
...
],
[
The X-coordinate of the points that make up the polygon,
The Y-coordinate of the points that make up the polygon,
...
],
...
],
"iscrowded" : 0 //not currently used
}, ...
],
"categories":[
{
"supercategory" : Object class supercategory (str),
"id" : Unique ID of the object class(int),
"name" : Name of the object class(str)
}, ...
]
}
Downloads
-
Images : train2025.zip val2025.zip test_open2025.zip
-
Annotations : instances_train2025.json instances_val2025.json
License
All datasets and benchmarks on this page are copyrighted by Hanwha Systems Co., and published under the Creative Commons Attribution-NonCommercial 4.0 International.