What is the TDWG Conference?
The annual TDWG conference is one of the major international events for the community working at the intersection of computer science and biodiversity. Organized every year by the TDWG association, it brings together researchers, collection managers, developers, data infrastructure managers, and representatives of national and international organizations. It aims to contribute to the development of current standards used to describe and work with biodiversity data.
But the conference is also a unique opportunity to present, discover, and discuss the latest scientific and technical advances in the field. The 2026 edition will take place in Oslo from September 21 to 25, 2026. This year’s event is being jointly organized by GBIF Norway, the Natural History Museum at the University of Oslo, and the Norwegian Biodiversity Information Centre. And if you can’t be there, don’t panic; there is a hybrid mode planned for the entirety of the event.
This year’s theme will be “Research and Robot-ready Biodiversity Data Standards”, highlighting the importance of data standards that can meet the needs of scientific research while also being directly usable by automated processing systems. These discussions will be particularly relevant in the current context of the rapid development of AI and digital infrastructures supporting research.
And this year, among the various talks, plenary sessions, and workshops, you will have several opportunities to hear from the Pl@ntNet team!
What will Pl@ntNet present during the Conference?
We will notably be taking part in Session SYM25: “From Data Mobilization to AI-ready Knowledge: Infrastructures for Multimodal Biodiversity Data.” This session will examine in detail the challenges involved in making use of the massive amounts of biodiversity data that are now available thanks to platforms such as Pl@ntNet. These platforms collect huge volumes of opportunistic observations, providing a valuable resource for ecological research and artificial intelligence.
Alexis Joly, Research Director at Inria and co-lead of Pl@ntNet, will give a presentation entitled “From 1.5 Billion Raw Queries to AI-Ready Biodiversity Data: Human–AI Collaborative Curation Pipelines in Pl@ntNet.” The presentation will describe how the more than 1.5 billion identification queries made through Pl@ntNet are progressively transformed into reliable and usable datasets. It will focus in particular on the collaborative curation pipeline developed by Pl@ntNet, which combines automated processing, community moderation, collaborative taxonomic review, image quality assessment, and governance requirements.
Alexis Joly will also present the various filters used to remove content that does not concern plants, low-quality observations, and problematic content, as well as the role of the community in identifying and reporting observations that are off-topic or considered unreliable.
The presentation will also demonstrate that the quality criteria used to assess these data are not necessarily the same depending on the intended use. The requirements of an AI model are not necessarily the same as those of a scientific infrastructure such as GBIF. The presentation will therefore provide an overview of how Pl@ntNet adapts its curation processes to meet these different objectives, while striving to make the data as reliable, reusable, and interoperable as possible.
A second presentation, this time given by Giulio Martelucci, a PhD researcher on the Pl@ntNet team, will focus on scaling challenges and new methods for improving Pl@ntNet’s performance while reducing processing times. Entitled “Scalable Edge AI for Real-Time Biodiversity Monitoring: A Case Study on Tracking Invasive Species in Roadside Imagery” the presentation will use the detection of invasive species in images taken along roads as a case study.
As part of his PhD research, Giulio is particularly interested in the limitations of the approach currently used by Pl@ntNet to analyze large images. These images are divided into numerous smaller areas, each of which requires multiple calls to the model to identify the species present. While this method produces accurate results, it involves a large number of queries and therefore becomes particularly costly in terms of time and computational resources as image volumes increase.
To address this issue, Giulio will explore and present an approach based on a lighter model, referred to as a “student” model, which can analyze the image as a whole and reproduce the results obtained by the “teacher” model currently used by Pl@ntNet Plots. The aim is to retain the richness of the results produced by the current method while significantly reducing the number of model calls required.
During his presentation, Giulio will discuss this approach and the methods used to enable the new model to achieve comparable or even better performance, while simplifying processing and reducing the load on the platform’s servers. In the longer term, this method could help make the analysis of large volumes of images significantly faster and more scalable, particularly for biodiversity monitoring and invasive species detection applications.
So, don’t hesitate to join us for these two presentations, as well as the many other talks taking place during the 2026 edition of TDWG. We hope to see you there!
