We are thrilled to share that the scientific paper “NoMeFormer: Non-Manifold Mesh Transformer” has won the Best Paper Award at the ISPRS Geospatial Week (April 2025) and has been published in the prestigious ISPRS Annals.
The paper, authored by our experts Mohammadreza Heidarianbaei and Franz Rottensteiner, together with Mareike Dorozynski and Max Mehltretter, introduces a groundbreaking approach to semantic segmentation of 3D meshes. It tackles one of the field’s most pressing challenges: processing non-manifold structures, a limitation that hinders most existing deep learning models.
To address this, the team developed NoMeFormer, a transformer-based framework capable of processing any type of mesh without structural constraints. Its core innovation, the Local-Global (L-G) transformer blocks, enables the model to capture long-range dependencies efficiently—without the high computational cost of conventional transformers.
This research represents a significant contribution to the field of AI and computer vision and is particularly important for the ChemiNova project, where complex, non-manifold mesh data is frequently encountered.
Congratulations to the entire team for their excellent work and outstanding achievements!
You can read the full paper here: https://isprs-annals.copernicus.org/articles/X-G-2025/365/2025/