AutoNaVIT-C : Vision-Based Path and Obstacle Segmentation Dataset for Autonomous Driving - XML Compatible

AutoNaVIT is a meticulously developed dataset designed to accelerate research in autonomous navigation, semantic scene understanding, and object segmentation through deep learning. This release includes only the annotation labels in XML format, aligned with high-resolution frames extracted from a controlled driving sequence at Vellore Institute of Technology – Chennai Campus (VIT-C). The corresponding images will be included in Version 2 of the dataset.

Class Annotations
The dataset features carefully annotated bounding boxes for the following three essential classes relevant to real-time navigation and path planning in autonomous vehicles:

Kerb – 1,377 instances

Obstacle – 258 instances

Path – 532 instances

All annotations were produced using Roboflow with human-verified precision, ensuring consistent, high-quality data that supports robust model development for urban and semi-urban scenarios.

Data Capture Specifications
The source video was captured using a Sony IMX890 sensor, under stable daylight lighting. Below are the capture parameters:

Sensor Size: 1/1.56", 50 MP

Lens: 6P optical configuration

Aperture: ƒ/1.8

Focal Length: 24mm equivalent

Pixel Size: 1.0 µm

Features: Optical Image Stabilization (OIS), PDAF autofocus

Video Duration: 4 minutes 11 seconds

Frame Rate: 2 FPS

Total Annotated Frames: 504

Format Compatibility and Model Support
AutoNaVIT annotations are provided in Pascal VOC-compatible XML format, making them directly usable with models that support the Pascal VOC standard. The dataset is immediately compatible with:

Pascal VOC

As XML is a structured, extensible format, these annotations can be easily adapted for use with additional object detection frameworks that support XML-based label schemas.

Benchmark Results
To assess dataset utility, a YOLOv8 segmentation model was trained on the full dataset (including images). The model achieved the following results:

Mean Average Precision (mAP): 96.5%

Precision: 92.2%

Recall: 94.4%

These metrics demonstrate the dataset’s effectiveness in training models for autonomous vehicle perception and obstacle detection.

Disclaimer and Attribution Requirement
By downloading or using this dataset, users agree to the terms outlined in the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License (CC BY-NC-ND 4.0):

This dataset is available solely for academic and non-commercial research purposes.

Proper attribution must be provided as follows:
“Dataset courtesy of Vellore Institute of Technology – Chennai Campus.”
This citation must appear in all research papers, presentations, or any work derived from this dataset.

Redistribution, public hosting, commercial use, or modification is prohibited without prior written permission from VIT-C.

Use of this dataset implies acceptance of these terms. All rights not explicitly granted are retained by VIT-C.