Dataset Card for FinnWoodlands

This is a FiftyOne dataset with 250 samples.

Installation

If you haven't already, install FiftyOne:

pip install -U fiftyone

Usage

import fiftyone as fo
from fiftyone.utils.huggingface import load_from_hub

# Load the dataset
# Note: other available arguments include 'max_samples', etc
dataset = load_from_hub("Voxel51/FinnWoodlands")

# Launch the App
session = fo.launch_app(dataset)

Dataset Card for FinnWoodlands

Dataset Details

Dataset Description

FinnWoodlands is a forest dataset designed for forestry and robotics applications. It consists of RGB stereo images, point clouds, and sparse depth maps, along with ground truth manual annotations for semantic, instance, and panoptic segmentation. The dataset comprises 4,226 manually annotated objects, with 2,562 objects (60.6%) corresponding to tree trunks classified into three instance categories: "Spruce Tree," "Birch Tree," and "Pine Tree." Additional annotations include "Obstacles" as instance objects and semantic "stuff" classes: "Lake," "Ground," and "Track."

• Curated by: Juan Lagos, Urho Lempiö, and Esa Rahtu (Tampere University, Finland)
• Shared by: Tampere University
• Language(s) (NLP): N/A (Computer Vision dataset)
• License: © Springer Nature Switzerland (exclusive license) - see paper for full terms

Dataset Sources

• Repository: https://github.com/juanb09111/FinnForest
• Paper: https://link.springer.com/chapter/10.1007/978-3-031-31435-3_7
• ArXiv: https://arxiv.org/abs/2304.00793
• Data Download (with annotations): https://drive.google.com/file/d/1uf9QBv1j_VRjM6jWCp2cgw5ZQ5CUx9R7/view
• Data Download (full, no GT): https://drive.google.com/drive/folders/1RhLxuHoxfB5C-Nz2_oyVAX1ima02Ddt8

Uses

Direct Use

• Semantic segmentation in forest environments
• Instance segmentation for tree trunk detection and species classification
• Panoptic segmentation for holistic forest scene understanding
• Depth completion from sparse depth maps
• Autonomous forestry robotics and navigation
• Development of data-driven methods for unstructured outdoor environments
• Benchmarking perception models for forest-like scenarios

Out-of-Scope Use

• Urban or indoor scene understanding (dataset is forest-specific)
• Tree species classification beyond the three included species (Spruce, Birch, Pine)
• Geographic generalization to forests outside Finland without domain adaptation
• Real-time applications without appropriate model optimization (benchmark models may require tuning)

Dataset Structure

Modalities:

• RGB stereo images

Annotation Types:

• Panoptic segmentation annotations

Classes:

Category Type	Class Name	Description
Things (Instance)	Spruce Tree	Tree trunk instances
Things (Instance)	Birch Tree	Tree trunk instances
Things (Instance)	Pine Tree	Tree trunk instances
Things (Instance)	Obstacles	Other countable objects
Stuff (Semantic)	Lake	Water bodies
Stuff (Semantic)	Ground	Forest floor
Stuff (Semantic)	Track	Forest paths/roads

Dataset Creation

Curation Rationale

Large and diverse datasets have driven breakthroughs in autonomous driving and indoor applications, but forestry applications lag behind due to a lack of suitable datasets. FinnWoodlands was created to address this gap and enable significant progress in developing data-driven methods for forest-like scenarios, particularly for applications requiring holistic environmental representation.

Source Data

Data Collection and Processing

Data was collected in Finnish forest environments using stereo camera setups capable of capturing RGB stereo images, point clouds, and sparse depth maps. The dataset captures unstructured forest scenarios that present unique challenges compared to urban or indoor environments.

Who are the source data producers?

Researchers at Tampere University, Finland, collected the data in Finnish woodland environments.

Annotations

Annotation process

Manual annotations were created using CVAT (Computer Vision Annotation Tool) for semantic, instance, and panoptic segmentation. The annotation scheme follows the "stuff" and "things" paradigm from computer vision, where "stuff" classes represent uncountable regions (Lake, Ground, Track) and "things" classes represent countable objects (tree trunks, obstacles).

Who are the annotators?

The research team at Tampere University performed the annotations.

Citation

BibTeX

@InProceedings{10.1007/978-3-031-31435-3_7,
author="Lagos, Juan
and Lempi{\"o}, Urho
and Rahtu, Esa",
editor="Gade, Rikke
and Felsberg, Michael
and K{\"a}m{\"a}r{\"a}inen, Joni-Kristian",
title="FinnWoodlands Dataset",
booktitle="Image Analysis",
year="2023",
publisher="Springer Nature Switzerland",
address="Cham",
pages="95--110",
}

APA

Lagos, J., Lempiö, U., & Rahtu, E. (2023). FinnWoodlands Dataset. In R. Gade, M. Felsberg, & J.-K. Kämäräinen (Eds.), Image Analysis (pp. 95-110). Springer Nature Switzerland. https://doi.org/10.1007/978-3-031-31435-3_7