Turin3D

license: mit language: - en tags: - urbanMapping - 3d - pointCloud - LiDAR pretty_name: Turin3D size_categories: - 10M<n<100M

Turin 3D Dataset

Description

The Turin 3D dataset is a collection of LiDAR point cloud data acquired within the city of Turin, Italy on January 2022 and collected in LAS 1.4 format. It's designed for use in 3D semantic segmentation tasks. This dataset offers a detailed 3D representation of the urban environment, enabling the development and evaluation of semantic segmentation models for urban scenes.

Purpose

This dataset is intended for researchers and practitioners interested in studying 3D semantic segmentation of urban environments using LiDAR data. It can be used for:

• Development and evaluation of 3D semantic segmentation algorithms.
• Analysis and understanding of urban scenes.
• Applications in autonomous vehicles, robotics, and urban planning.

Dataset Content

It contains almost 70M points divided among 57 blocks covering around 25k m^2 each. Data is compressed with gzip, you can extract it using

cat dataset.tar.*.gz.part > dataset.tar.gz
tar -xvzf dataset.tar.gz

Class Taxonomy

The dataset utilizes a taxonomy of 6 semantic classes:

1. Undefined
2. Soil
3. Terrain
4. Vegetation
5. Building
6. Street element
7. Water

Dataset Splits

• Train: Provided with soft labels (class probabilities).
• Validation: Precise annotations (hard labels) for model evaluation.
• Test: Precise annotations (hard labels) for final model evaluation.

Annotation

• Train: Annotations generated automatically using deep learning models, providing soft labels.
• Validation and Test: Manual annotations.

Dataset Card Authors

Luca Barco, Giacomo Blanco, Gaetano Chiriaco, Fabrizio Dominici

Dataset Card Contact

luca.barco@linksfoundation.com, giacomo.blanco@linksfoundation.com, gaetano.chiriaco@linksfoundation.com, fabrizio.dominici@linksfoundation.com