Jul 2026· 2026 IEEE International Workshop on Metrology for Living Environment (MetroLivEnv)· pp. 200-205· 0 citations· 16 references
Abstract
The automated detection and classification of surface degradation on historical buildings represents a critical challenge in architectural heritage conservation. Conventional approaches relying on manual inspection or supervised machine learning require extensive annotated datasets and expert involvement, limiting their scalability. This paper presents a novel zero-shot pipeline for degradation segmentation on historical civil architecture, combining UAV-acquired photogrammetric data processed in Agisoft Metashape with Gemma 4 31B, Google DeepMind's flagship open-weight vision language model, running locally via LM Studio, and the Segment Anything Model 2 (SAM2) for pixel-accurate mask generation. The system operates entirely without task-specific training data, producing segmentation masks overlaid on the RGB orthomosaic for expert visual evaluation. A case study on a degraded historical building in Calabria, southern Italy, demonstrates the pipeline's ability to detect and categorize detachment, cracking, and lacunae in a unified, reproducible workflow. Results are evaluated through structured expert visual assessment. The approach offers a replicable, low-cost alternative to supervised segmentation, particularly suited to contexts where labeled data is unavailable.
The first systematic zero-shot evaluation of SAM 2 for aerial building segmentation is presented, establishing SAM 2 as a viable tool for rapid building mapping while highlighting where domain adaptation remains necessary.
Bingning Xiong, Mingyu Ou· Journal of image processing...· 0 citations
Abstract. Accurate damage localization is essential for infrastructure inspection, but conventional segmentation methods rely on dense pixel-level annotations that are costly to obtain and difficult to scale. This paper presents a foundation model-based pipeline for data-efficient damage localization in built infrastructure. The proposed workflow combines DINOv3 features for image-level classification, Grad-CAM for weak localization, and the Segment Anything Model (SAM) for prompt-guided pixel-level segmentation. The resulting masks are further transferred into 3D space for spatially contextualized visualization. The pipeline is evaluated on two case studies. On a subset of Sewer-ML, three representative sewer defect classes are used to compare pretrained backbones and to qualitatively assess downstream localization. The DINOv3-based classifier achieves a higher average F2-score than a Google ViT baseline, reaching about 0.72 versus 0.64. On a custom historic masonry dataset, the method is quantitatively evaluated for material-loss segmentation using manually annotated test masks. The proposed heatmap-guided prompting strategy achieves a mean Dice score of 0.69 and a mean IoU of 0.53, while the classification stage reaches an F2-score of 0.99. A proof-of-concept experiment further demonstrates that segmented damage regions can be visualized within a larger local 3D scene. Overall, the results show that the proposed foundation-model based pipeline can support data-efficient and spatially meaningful damage localization across different infrastructure domains.
Zhi-Ya Yang, Roberto de Lima-Hernandez, M. Vergauwen· The International Archives o...· 0 citations
Abstract. This study investigates the enrichment of existing building models using deep learning-based window detection from oblique aerial imagery acquired by a high-end multi-camera sensor system. While many cities maintain LOD2 building models at Level of Detail 2, higher levels of detail require the integration of facade elements such as windows. Three detection strategies are evaluated using 3D reference building models to assess accuracy and completeness. The test site is located in Vienna and consists of multiple large residential buildings with varying facade characteristics. The evaluated methods include zero-shot object detection with Grounding DINO combined with Segment Anything Model 2, applied to both oblique images and facade orthophotos, as well as a SAM2-UNeXT network requiring minimal training. Results indicate that zero-shot detection on orthophotos achieves the best performance, with a precision of 0.95 and an F1 score of 0.85. In contrast, the SAM2-UNeXT approach shows lower precision and F1 scores but slightly higher recall. The investigation shows that detection performance is influenced by facade viewing angles. Steeper viewing angles generally improve detection quality but increase susceptibility to occlusions, particularly in dense urban environments. The article concludes with a detailed outlook on future work, including the extension of the approach to more complex three-dimensional building structures.
J. Otepka, Günter Sükar, Martin Kerschner et al.· The International Archives o...· 0 citations
In the present generation of increasing geospatial data, accurate and automated extraction of building footprints from high-resolution aerial and satellite imagery has become crucial for various applications such as urban planning, infrastructure development, disaster management, and GIS database maintenance, as manual tracing is time-consuming and unstable for large-scale mapping. This study compares conventional image processing techniques such as thresholding, edge detection, morphological operations through a machine learning approach using Random Forest (RF), and deep learning-based semantic segmentation models, namely U-Net and DeepLabV3+, along with the Segment Anything Model (SAM) using a pre-trained prompt-based setup. All methods are tested on the same set of data, and a standardized data preprocessing is performed for fair comparison. The overall results indicate that the application of DeepLabV3+ is best, with an IoU of 82% and an F1 score of 90%. U-Net achieves second high IoU and F1 scores of 74% and 84% respectively, while Random Forest shows a high IoU of 60% and an F1-score of 72%. SAM has the lowest scores with an IoU of 50% and an F1 score of 51%.
Experimental results demonstrate that the adapted SAM2 model achieves stable segmentation under moderate environmental variability, while degrading under severe visibility loss, consistent across model scales and input resolutions.
Bindusara Nagathihalli Lokesh, Laura Camila Duran Vergara, Hans-Gerd Maas et al.· The International Archives o...· 1 citation
Routine visual inspections of bridges are safety-critical activities that are still manual, time-consuming, and subjective to the inspector’s interpretation. This paper presents a scalable automated visual inspection pipeline for semantic segmentation of bridge defects, targeting the 19-class dacl10k benchmark dataset. We perform the first systematic comparison of pretraining paradigms on dacl10k, including CNN -supervised, ViT-supervised, masked image modelling, and self-supervised learning across 10 configurations under a controlled training protocol. The pretraining paradigm consistently dominates architectural choice, with DINOv2-L, pretrained on 142 million unlabelled images, achieving a mean Intersection-over-Union (mIoU) of 49.16%. Applying our native multi-label training approach to EVA-02-L, the dacl10k challenge-winning backbone, achieves 48.97% mIoU versus their 47.80% single-model result, demonstrating that training design is an independent performance factor. A three-model ensemble achieves 51.08% mIoU, exceeding the challenge winner’s score by using half the number of models. A prototype inspection system (InSpectralytiX) is deployed in a HuggingFace Gradio Space, demonstrating end-to-end feasibility from raw image to a per-class defect map. The future work targets automated condition scoring for bridge asset management integration, supporting structural health monitoring at the local level when performed repeatedly.
R. Khan, R. Kromanis· e-Journal of Nondestructive...· 0 citations