You need to agree to share your Hugging Face information to access this model
This repository is publicly accessible, but you have to accept the conditions to access its files and content.
TITAAN-SD-yearly is shared for early testing only. To request access, contact datascience@hetwaterschapshuis.nl.
Log in or Sign Up to review the conditions and access this model content.
TITAAN-SD-yearly
Early testing only. This model is shared with a small group of partners to try it out. It is not a final release: it has no licence yet, and the weights and documentation may change. To request access, contact datascience@hetwaterschapshuis.nl.
TITAAN-SD-yearly is a vision encoder for Dutch aerial imagery. It starts from Meta's DINOv2 (ViT-L/14 with registers) and was trained further, without labels, on yearly 8 cm aerial photographs of the Netherlands together with color-infrared imagery and AHN heights. For every 1.28 m of ground it produces a 1024-number embedding that downstream models can use, for example to map water, buildings, jetties or boats.
The repository contains the encoder in Hugging Face Transformers format. After access is granted, DETAILS.md
describes the inputs, how to compute embeddings, and early evaluation results. The DINOv2 licence for the base
model is included as LICENSE-DINOv2.
- Downloads last month
- -
Model tree for hwh-datascience/TITAAN-SD-yearly
Base model
facebook/dinov2-with-registers-large