FCDD: A High-Resolution Unstructured Environment Dataset with Multiple Sand Roads
Description
The Floripa Coast Driving Dataset (FCDD) is a 4K-resolution semantic segmentation dataset designed for autonomous driving and unstructured environment perception. It covers a 12.2 km coastal route featuring asphalt, dirt roads, and multiple sand terrains (soft sand, hard sand, and wet sand), captured under diverse weather and lighting conditions. FCDD comprises 500 pixel-level annotated images (400 for training / 100 for validation) across 17 semantic classes, including road surfaces, markings, cat's eyes, cracks, potholes, vehicles, people, animals, and obstacles. It supports research on terrain classification, road distress detection, and coastal navigation, and enables studies on texture recognition, multi-surface segmentation, and the impact of image resolution. See the original article: "FCDD: A High-Resolution Unstructured Environment Dataset with Multiple Sand Roads." at IEEE Access.
Files
Steps to reproduce
1. Data Acquisition - Record with a GoPro HERO11 across three sessions under varied weather, lighting, and seasons. - Capture a 12.2 km route (27.5689°S, 48.4343°W → 27.4817°S, 48.3818°W) including asphalt, dirt, and sand (soft, hard, wet) roads. 2. Frame Extraction - Extract 4K frames (3840×2160 px) and select 500 diverse images. - Split into 400 training and 100 validation samples from non-overlapping video segments. 3. Annotation - Label pixel-wise in Roboflow using 17 classes: - Surfaces (asphalt, dirt, hard/soft/wet sand), - Signs (markings, cat’s eyes, walls, bumps), - Distress (cracks, potholes), - Things (ego, vehicles, people, animals, obstacles). - Use background as a void class. 4. Class Definition - Soft sand: dry, non-drivable. - Hard sand: dry, compact, drivable. - Wet sand: moist, firm, drivable. - Label only drivable or relevant road areas; all else as background.
Institutions
- Universidade Federal de Santa Catarina