Yamaha-CMU Off-Road Dataset (YCOR)
CMU AirLab · 2018 · datasets.bot · datasets.bot page
One-liner. 1,076 off-road RGB images from four sites in Western Pennsylvania and Ohio across three seasons, pixel-labeled into 8 terrain/traversability classes for off-road semantic segmentation.
Setup
- Datasets / benchmarks: The Yamaha-CMU Off-Road (YCOR) dataset is a semantic segmentation benchmark for autonomous off-road navigation, released by CMU AirLab. It contains 1,076 egocentric RGB images collected from a Yamaha off-road vehicle at four locations in Western Pennsylvania and Ohio, spanning three different seasons to capture varied terrain and lighting conditions. Each image is densely pixel-labeled into 8 classes (sky, rough trail, smooth trail, traversable grass, high vegetation, non-traversable low vegetation, obstacle, and puddle) to support terrain traversability estimation. Labels were created with a polygon-based annotation interface, densified via Dense CRF post-processing, then manually inspected and corrected. The dataset is split into 931 training and 145 validation images, with the split generated so that no data-collection session overlaps between train and validation. It accompanies the 2018 Field and Service Robotics paper 'Real-Time Semantic Mapping for Autonomous Off-Road Navigation' by Maturana, Chou, Uenoyama, and Scherer, and is widely used for off-road semantic segmentation and traversability research. License: CC-BY-4.0. Download: https://cmu.box.com/s/3fngoljhcwhqf2z5cbepufh331qtesxt.
- Hardware / simulator: Embodiment: not listed. Environment: outdoor. Realness: physical.
Schema
1,076 RGB off-road images with per-pixel semantic segmentation masks over 8 terrain/traversability classes (sky, rough trail, smooth trail, traversable grass, high vegetation, non-traversable low vegetation, obstacle, puddle); 931 train / 145 validation split.
Links