Dataset Specification

The dataset is consisted of HD map dataset & localization dataset

App App
HD Map Dataset & Localization Dataset

NAVER LABS mapping technology is based on the integration of city-scale aerial photographs with data from a mobile mapping system. We extract information about the layout of the road surface (3D Road Layout) from aerial images. Then we integrate it with 3D point cloud collected by R1, our lightweight mobile mapping system (MMS). Compared to conventional HD maps constructed by MMS vehicles, our mapping process can significantly reduce the production costs and time. We also provide a separate localization dataset containing sensor data and pseudo ground truth poses which can be used for evaluation purposes.

HD Map Dataset

Our HD map dataset consists of the following three components:

3D Road Layout


We provide the 3D layout of the road surface that we extract from aerial photographs. It contains information regarding the types and precise 3D locations of visual structures on the road surface that are essential for self-driving vehicles, such as lanes, road markings, crosswalks, crossroads and speed bumps.

LiDAR Feature Data


Our HD map dataset also includes 3D LiDAR point cloud of the surrounding road environment captured from our MMS vehicle. Each point is associated with a semantic label indicating the type of object or region it belongs to. Dynamic objects like vehicles and pedestrians are automatically detected and removed from the point cloud.

Visual Feature Data


Final component of our HD map dataset is a set of visual features extracted from salient regions of the road environment. Obtained by a deep learning model trained in-house, these features are compact, discriminative, and invariant across different viewing conditions, allowing for reliable matching and localization.

Data File Type Dimension Classes QuadTile Size Path Format
3D Road Layout GeoJSON [x,y,z, type] Lane, Stop, Link, Sign, ... -
LiDAR Feature Data LAS [x,y,z, intensity, classification] Road / Pole / Other 20.48m x 20.48m x Areal UTM
Visual Feature Data H5 [x,y,z, roll, pitch, yaw, feature_descriptor] 128D Feature Descriptor 20.48m x 20.48m x Areal UTM

Localization Dataset

It is important to accurately estimate the current location of the vehicle in autonomous driving situations. The localization technology of NAVER LABS utilizes diverse sensors such as LiDAR, Cameras, GPS, IMU and Wheel Encoders. The localization dataset provides a collection of raw data from these sensors and corresponding pseudo ground truth poses for the purpose of evaluating your own localization algorithms.

App App App

File Structure

Meta Data Sensor Data
Calibration Camera 3D / 2D-LiDAR GPS / INS FOG OBD Wheel Encoder