AutoNaVIT : Vision-Based Path and Obstacle Segmentation Dataset for Autonomous Driving - TXT Compatible
Description
AutoNaVIT is a meticulously curated dataset developed to assist research in autonomous navigation, scene understanding, and deep learning-based object segmentation. This release contains only the annotation labels in TXT format corresponding to high-resolution frames extracted from a recorded driving sequence at Vellore Institute of Technology – Chennai Campus (VIT-C). The corresponding images will be made available in Version 2 of the dataset soon. The dataset features manually annotated bounding boxes and labels for three essential classes critical for autonomous vehicle navigation: Kerb – 1,377 instances Obstacle – 258 instances Path – 532 instances All annotations were created using Roboflow, ensuring high fidelity and consistency, which is vital for real-world autonomous driving applications in both urban and semi-urban environments. Data Capture Specifications Source imagery was recorded using a Sony IMX890 sensor with the following specifications: Sensor Size: 1/1.56", 50 MP Lens: 6P, ƒ/1.8, 24mm equivalent, 1.0 µm pixels Features: OIS (Optical Image Stabilization), PDAF autofocus Video Duration: 4 min 11 sec Frame Rate: 2 FPS Total Annotated Frames: 504 Format Compatibility and Model Support AutoNaVIT annotations are provided in standard TXT format, enabling direct compatibility with the following 13 models: yolokeras yolov4pytorch darknet yolov5-obb yolov8-obb imt-yolov6 yolov4scaled yolov5pytorch yolov7pytorch yolov8 yolov9 yolov11 yolov12 As the dataset adheres to standard YOLO TXT annotations, it can easily be adapted for other models or frameworks that support TXT-based annotations. Benchmark Results To evaluate the dataset’s performance, a YOLOv8-based segmentation model was trained on the complete dataset (images + annotations). The model achieved: Mean Average Precision (mAP): 96.5% Precision: 92.2% Recall: 94.4% These results confirm the dataset's high utility and reliability in training segmentation models for autonomous vehicle perception systems. Disclaimer and Attribution Requirement By accessing or using this dataset, users agree to the terms outlined under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License (CC BY-NC-ND 4.0): Usage is permitted solely for non-commercial academic and research purposes. Proper attribution must be given, stating: “Dataset courtesy of Vellore Institute of Technology – Chennai Campus.” This acknowledgment must be included in all forms of publication, presentation, or dissemination of work utilizing this dataset. Redistribution, commercial use, modification, or public hosting of the dataset is prohibited without explicit written permission from VIT-C. Use of this dataset implies acceptance of these terms. All rights not explicitly granted are reserved by VIT-C.
Files
Steps to reproduce
The following steps outline the process to recreate AutoNaVIT: Vision-Based Path and Obstacle Segmentation Dataset for Autonomous Driving – TXT Compatible, which provides segmentation labels in plain .txt format. The dataset is designed to assist researchers and engineers working on object segmentation for autonomous navigation and is directly compatible with 13 YOLO-based models. The corresponding image dataset will be released in Version 2. 1. Data Collection Begin by recording a video along a predefined path using a high-resolution camera. For best results, a sensor with specifications similar to the Sony IMX890 is recommended. Key features include: 50MP resolution 24mm focal length ƒ/1.8 aperture Optical Image Stabilization (OIS) Capture footage under daylight conditions at a standard frame rate (preferably 30 FPS) to ensure clarity and consistency. 2. Frame Extraction Extract frames from the recorded video at a rate of 2 frames per second (FPS). This offers a sufficient variety of scenes while avoiding redundancy. Maintain the original resolution of the frames for accurate annotation. 3. Annotation Using Roboflow Upload the extracted frames to Roboflow or a similar annotation tool to label objects using bounding boxes or polygon-based masks, depending on the target model's compatibility. Define the following three classes: Kerb Obstacle Path After accurate manual annotation, export the dataset in YOLO-compatible TXT format, where each .txt file corresponds to an image and contains label class ID, bounding box coordinates, and normalized values. 4. Model Training and Evaluation To evaluate the dataset’s quality, train a segmentation model using any supported YOLO architecture. In the official benchmark using YOLOv8, the dataset achieved the following: Mean Average Precision (mAP): 96.5% Precision: 92.2% Recall: 94.4% These metrics confirm the dataset’s strength in supporting robust segmentation models for autonomous driving. 5. Format Compatibility and Model Support The annotations are provided in a universal TXT format compatible with standard YOLO configurations. AutoNaVIT is ready to be used with the following 13 models: yolokeras yolov4pytorch darknet yolov5-obb yolov8-obb imt-yolov6 yolov4scaled yolov5pytorch yolov7pytorch yolov8 yolov9 yolov11 yolov12 Because of its standard TXT formatting, the dataset can be further modified to fit the label structure of any other model that supports YOLO-format TXT annotations. By following these steps, users can reproduce the AutoNaVIT dataset, train compatible object segmentation models, and leverage the data for research and development in autonomous navigation systems.