Figure 3 - available via license: Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International
Content may be subject to copyright.
Download
View publication
CLIP model architecture (Radford et al., 2021)

CLIP model architecture (Radford et al., 2021)

Source publication
Figure 1. Selected study areas from the ISPRS Vaihingen dataset
Figure 2. Segment anything model architecture (Kirillov et al., 2023)
Figure 3. CLIP model architecture (Radford et al., 2021)
Figure 4. Detection of roof surfaces from aerial photographs and...
+4 Figure 5. Detection of roof surfaces from point clouds and calculation...
Vision Foundation Models and Rule-Based Approaches for Roof Surface Segmentation and Photovoltaic Potential Analysis in Urban Areas
Article
Full-text available
  • Mar 2025
This study presents two methods for rapidly and effectively determining the photovoltaic (PV) potential of building roofs in urban areas using aerial photographs and point cloud data. In the first method, the Segment Anything Model (SAM) and Contrastive Language Image Pre-Training (CLIP) models are used to detect roof surfaces and obstacles from ae...
Cite
Download full-text

Context in source publication

Context 1
... CLIP model can perform image-text matching using instructions given in natural language. The architecture of CLIP utilizes two separate networks to process language and image data: a Vision Transformer (ViT) model and a language model (a Transformer-based text encoder) (Figure 3). CLIP enhances the learning of semantic congruence between images and associated text by bringing the image and its related text closer together. ...