Daan de Geus

I'm an assistant professor in the Mobile Perception Systems Lab at the Eindhoven University of Technology (TU/e), in the Netherlands. I did my PhD at TU/e, and was a visiting PostDoc in the Computer Vision Lab at RWTH Aachen University, headed by Prof. Dr. Bastian Leibe. My research focuses on machine learning for visual and multiomodal scene understanding.

Scholar / Twitter / Bluesky / Github

News

[2025/07] I started as an assistant professor at the MPS Lab of TU/e.
[2025/06] DONUT has been accepted to ICCV 2025.
[2025/05] We have won the CVPR 2025 ScanNet++ Challenge with DITR.
[2025/03] EoMT has been accepted to CVPR 2025.

Selected publications

See all publications here.

	DONUT: A Decoder-Only Model for Trajectory Prediction Markus Knoche, Daan de Geus, Bastian Leibe ICCV, 2025 Project page / arXiv / Code (soon) We predict future trajectories autoregressively with a decoder-only model, to treat historical and future trajectories identically.
	Your ViT is Secretly an Image Segmentation Model Tommie Kerssies, Niccolò Cavagnero, Alexander Hermans, Narges Norouzi, Giuseppe Averta, Bastian Leibe, Gijs Dubbelman, Daan de Geus CVPR, 2025 (Highlight Paper) Project page / arXiv / CVF / Code With a sufficiently large model and extensive pre-training, complex task-specific components are not necessary for image segmentation.
	DINO in the Room: Leveraging 2D Foundation Models for 3D Segmentation Karim Abou Zeid, Kadir Yilmaz, Daan de Geus, Alexander Hermans, David Adrian, Timm Linder, Bastian Leibe arXiv, 2025 Project page / arXiv / Code (soon) We inject image-based DINOv2 features into a point cloud model to dramatically enhance 3D segmentation performance.
	Fine-Tuning Image-Conditional Diffusion Models is Easier than You Think Gonzalo Martin Garcia, Karim Abou Zeid, Christian Schmidt, Daan de Geus, Alexander Hermans, Bastian Leibe WACV, 2025 (Oral Presentation)* Project page / arXiv / CVF / Code / Demo Repurposing diffusion models for geometry estimation is as simple as end-to-end fine-tuning.
	ALGM: Adaptive Local-then-Global Token Merging for Efficient Semantic Segmentation with Plain Vision Transformers Narges Norouzi, Svetlana Orlova, Daan de Geus, Gijs Dubbelman CVPR, 2024 Project page / arXiv / CVF / Code By merging patch tokens locally and then globally, the throughput of ViT-based segmentation models can be greatly enhanced while preserving accuracy.
	Content-aware Token Sharing for Efficient Semantic Segmentation with Vision Transformers Chenyang Lu, Daan de Geus*, Gijs Dubbelman CVPR*, 2023 Project page / arXiv / CVF / Code A small pre-processing network identifies image patches that can share a token in a ViT-based segmentation model, to improve efficiency without harming the accuracy.

Based on Jon Barron's website.