DeepTrees Practitioners Workshop: Tree Crown Segmentation and Analysis in Remote Sensing Imagery with PyTorch

Tuesday 25 Nov 2025, 09:00 → 17:15 Europe/Berlin

Hall 1 C (Leipziger KUBUS, Permoserstraße 15, 04318 Leipzig)

Taimur Khan (Helmholtz-UFZ)

Summary

The DeepTrees Practitioners Workshop 2025 places the DeepTrees Python library (https://deeptrees.de) at the heart of our efforts to democratize tree crown segmentation and analysis. With over 6,500 downloads on PyPI, DeepTrees offers an end-to-end, modular PyTorch framework for semantic and instance segmentation of tree crowns, integrated with tools for ecological trait extraction and active learning.   This workshop invites researchers, environmental data scientists, and remote sensing practitioners to deepen their expertise using this open-source tool. Participants will engage with the full workflow — from dataset preparation and annotation, to model training, evaluation, and trait extraction — while also hearing flash updates from the open-source community about new features, community contributions, and evolving use cases. Through hands-on sessions and collaborative dialogue, we aim to empower attendees to adopt, extend, and contribute to DeepTrees, fostering a vibrant ecosystem where practitioners, researchers, and data custodians converge to accelerate scalable, reproducible tree monitoring across landscapes.

Who Should Attend

• Machine learning practitioners and data scientists in environmental domains
• Researchers in remote sensing, forestry, and ecology
• GIS analysts and urban planners using high-resolution imagery
• Students and developers interested in PyTorch and spatial AI

Requirements

• Some experience with Python programming.
• Understanding of basic machine learining concepts.
• (OPTIONAL) Some knowledge of neural networks.

 

Participants must bring a laptop for participation in the hands-on sessions.

1 Welcome & Introduction to DeepTrees

Overview of project goals, workshop aims, and roadmap.

2 Flash Talks: Open Source & Package Updates

Short (~15 min) lightning presentations from the community on:
- New features or modules in deeptrees
- Use cases or applications built on deeptrees
- Community datasets published or curated
- Challenges, contributions, and future plans

10:45 Coffee break

3 Hands-on Session 1: Fundamentals of Tree Crown Segmentation with Deep Learning

Introduction to convolutional neural networks, UNet architectures, and the DeepTrees adaptation for tree crown delineation.
Hands-on: Model architecture overview in PyTorch using sample orthoimages.

12:00 Lunch break

4 Hands-on Session 2: Data Annotation and Preprocessing

Working with aerial and UAV imagery — data formats, tiling, annotation (QGIS), and mask generation.
Hands-on: Creating training and validation datasets.

5 Hands-on Session 3: Training and Improving a Tree Crown Segmentation Model with Active Learning

Step-by-step coding walkthrough for training and fine-tuning a DeepTrees segmentation model.
Hands-on: Participants train and evaluate their own models on provided data.

15:00 Coffee break

6 Hands-On Session 4: Model Evaluation and Ecological Analysis

Metrics for segmentation performance (IoU, precision-recall, F1 score) and post-processing for ecological insights (e.g., canopy cover, crown radius, species inference).
Hands-on: Interpreting results and visualizing segmented tree crowns.

7 Panel and Open Discussion: From Models to Monitoring Systems

Discussion on integrating outputs into DeepTrees workflows, FAIR data publication, and collaborative opportunities for long-term ecological monitoring