Automated LoD2 Building Reconstruction Using Bimodal Segmentation
Bibliographic record
Abstract
LoD2 3D modeling of buildings has rapidly developed, especially in digital twin and urban planning applications. However, the diversity of roof structures makes significant challenges for 3D building modeling. While Airborne images are a key data source for capturing optical roof features, they often struggle in low contrast areas or shadowed conditions. LiDAR data can provide height information to compensate for the occluded areas. However, in current literature, particularly in existing deep learning models, the integration of these bimodal data has not been effectively studied for segmentation and modeling. This study proposes a novel multi-object instance segmentation framework that jointly detects all geometrical roof components required for 3D building modeling, such as planes, inlines, and outlines, in a single bimodal network. Our method uses a shared ResNet-50 backbone with Feature Pyramid Networks (FPN) and introduces two custom attention mechanisms: Inter-Modality Attention Blocks (IMAB) for fusing elevation and spectral features, and Modality-Specific Attention Blocks (MSAB) for refining modality-specific information. We also evaluate three fusion strategies, early, middle, and late, to optimize bimodal integration. Extracted lines and planes are vectorized, combined with height data from point clouds, and used to create 3D planes and lines. The final building models and wireframes are constructed by intersecting these 3D components. To the best of our knowledge, this is the first study to combine bimodal multi-object instance segmentation with attention mechanisms to jointly extract all roof components and create 3D building models. Our study demonstrates that combining the optical and elevation features of each building improves the accuracy of building geometric component segmentation and 3D building modeling. Experiments conducted with 1,488 buildings, show an F2-score of 0.8620, F1-score of 0.8374, and IoU of 0.844 for the segmentation step, and an RMSE of around one meter for the 3D modeling.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.001 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.002 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.004 | 0.002 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".