Cost-Effective High-Definition Building Mapping: Box-Supervised Rooftop Delineation Using High- Resolution Remote Sensing Imagery
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Deep learning–based high-definition building mapping faces challenges due to the need for extensive high-quality training data, leading to significant annotation costs. To mitigate this challenge, we introduce Box2Boundary, a novel approach using box supervision, in conjunction with the segment anything model (SAM), to achieve cost-effective rooftop delineation. Leveraging the tiny InternImage architecture for enhanced feature extraction and using the dynamic scale training strategy to tackle scale variance, Box2Boundary demonstrates superior performance compared to alternative box-supervised methods. Extensive experiments on the Wuhan University Building Data Set validate our method's effectiveness, showcasing remarkable results with an average precision of 48.7%, outperforming DiscoBox, BoxInst, and Box2Mask by 22.0%, 11.3%, and 2.0%, respectively. In semantic segmentation, our method achieved an F 1 score of 89.54%, an overall accuracy (OA) of 97.73%, and an intersection over union (IoU) of 81.06%, outperforming all other bounding-box-supervised methods, image tag–supervised methods, and most scribble-supervised methods. It also demonstrated competitive performance compared to fully supervised methods and scribble-supervised methods. SAM integration further boosts performance, yielding an F 1 score of 90.55%, OA of 97.84%, and IoU of 82.73%. Our approach's efficacy extends to the Waterloo Building and xBD Data Sets, achieving an OA of 98.48%, IoU of 84.72%, and F 1 score of 91.73% for the former and an OA of 97.32%, IoU of 60.10%, and F 1 score of 75.08% for the latter. These results underscore the method's robustness and cost-effectiveness in rooftop delineation across diverse data sets.
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.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.001 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.001 | 0.006 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
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 it