CCU-Truck Dataset

Dataset Overview

CCU-Truck is a scene-centric vehicle detection dataset collected from real-world traffic scenes in Taiwan. The dataset is designed for fine-grained truck and vehicle subclass detection, focusing on both common vehicles and special-purpose trucks that frequently appear in highway and road surveillance scenarios.

The dataset contains 1,083 annotated images with bounding-box labels for 12 vehicle categories. The images include complex traffic environments, different object scales, partial occlusions, diverse viewpoints, and visually similar vehicle subclasses, making the dataset suitable for evaluating object detection models under realistic traffic conditions.

Dataset Information

• Dataset Name: CCU-Truck
• Task Type: Object Detection
• Number of Images: 1,083
• Number of Classes: 12
• Annotation Format: YOLO format
• Image Format: JPG
• Application: Fine-grained vehicle and truck subclass detection

Annotation Classes

The dataset contains 12 classes:

• Class 0: box_truck
• Class 1: bus
• Class 2: car
• Class 3: concrete_mixer_truck
• Class 4: dump_truck
• Class 5: flatbed_truck
• Class 6: impact_attenuator_truck
• Class 7: mini_truck
• Class 8: mobile_crane
• Class 9: tanker_truck
• Class 10: tow_truck
• Class 11: tractor_unit

Annotation file format: text class_id x_center y_center width height

All bounding-box coordinates are normalized in YOLO format.

Annotation Distribution

Class	Train	Valid	Test	Total Annotations	Percentage
car	1,602	555	326	2,483	53.58%
box_truck	519	177	86	782	16.88%
mini_truck	248	73	41	362	7.81%
flatbed_truck	160	50	32	242	5.22%
tanker_truck	139	44	29	212	4.57%
dump_truck	135	49	23	207	4.47%
bus	136	42	18	196	4.23%
mobile_crane	41	9	6	56	1.21%
impact_attenuator_truck	17	16	5	38	0.82%
concrete_mixer_truck	6	3	12	21	0.45%
tow_truck	10	5	6	21	0.45%
tractor_unit	6	4	4	14	0.30%

Split-level Annotation Counts

• Training annotations: 3,019
• Validation annotations: 1,027
• Test annotations: 588
• Total annotations: 4,634

Dataset Structure

CCU-Truck/
├── train/
│   ├── images/        # Training image files
│   └── labels/        # YOLO-format training labels
├── valid/
│   ├── images/        # Validation image files
│   └── labels/        # YOLO-format validation labels
├── test/
│   ├── images/        # Test image files
│   └── labels/        # YOLO-format test labels
└── README.md

Usage Instructions

1. The dataset can be downloaded from Roboflow.
2. The dataset supports YOLO-format object detection training.
3. The dataset can be used for training and evaluating fine-grained vehicle detection models.
4. The validation and test splits can be used to evaluate detection performance on scene-centric traffic images.