Papers › Deep learning approaches to building rooftop thermal bridge detection from aerial images

Deep learning approaches to building rooftop thermal bridge detection from aerial images

12 Dec 2022Automation in Construction 2022 12archive 2025-07-28

Zoe Mayer, James Kahn, Yu Hou, Markus Götz, Rebekka Volk, Frank Schultmann

Thermal bridges are weak points of building envelopes that can lead to energy losses, collection of moisture, and formation of mould in the building fabric. To detect thermal bridges of large building stocks, drones with thermographic cameras can be used. As the manual analysis of comprehensive image datasets is very time-consuming, we investigate deep learning approaches for its automation. For this, we focus on thermal bridges on building rooftops recorded in panorama drone images from our updated dataset of Thermal Bridges on Building Rooftops (TBBRv2), containing 926 images with 6,927 annotations. The images include RGB, thermal, and height information. We compare state-of-the-art models with and without pretraining from five different neural network architectures: MaskRCNN R50, Swin-T transformer, TridentNet, FSAF, and a MaskRCNN R18 baseline. We find promising results, especially for pretrained models, scoring an Average Recall above 50% for detecting large thermal bridges with a pretrained Swin-T Transformer model.

PaperPDFCode

Code

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Instance SegmentationObject Detection

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Instance Segmentation TBBR Swin-T (ImageNet-1k pretrain) Average Recall@IoU:0.5-0.95 28.0 #1 of 5 Archive leaderboard report
Instance Segmentation TBBR Mask R-CNN (ResNet-50-FPN, ImageNet-1k pretrain) Average Recall@IoU:0.5-0.95 21.9 #2 of 5 Archive leaderboard report
Instance Segmentation TBBR Swin-T Average Recall@IoU:0.5-0.95 20.6 #3 of 5 Archive leaderboard report
Instance Segmentation TBBR Mask R-CNN (ResNet-50-FPN) Average Recall@IoU:0.5-0.95 20.1 #4 of 5 Archive leaderboard report
Object Detection TBBR Swin-T (ImageNet-1k pretrain) Average Recall@IoU:0.5-0.95 45.4 #1 of 7 Archive leaderboard report
Object Detection TBBR FSAF (ResNeXt-101, ImageNet-1k pretrain) Average Recall@IoU:0.5-0.95 38.0 #2 of 7 Archive leaderboard report
Object Detection TBBR Mask R-CNN (ResNet-50-FPN, ImageNet-1k pretrain) Average Recall@IoU:0.5-0.95 37.0 #3 of 7 Archive leaderboard report
Object Detection TBBR Mask R-CNN (ResNet-50-FPN) Average Recall@IoU:0.5-0.95 30.8 #4 of 7 Archive leaderboard report
Object Detection TBBR TridentNet (ResNet-50, ImageNet-1k pretrain) Average Recall@IoU:0.5-0.95 30.0 #5 of 7 Archive leaderboard report
Object Detection TBBR FSAF (ResNeXt-101) Average Recall@IoU:0.5-0.95 24.8 #6 of 7 Archive leaderboard report
Object Detection TBBR TridentNet (ResNet-50) Average Recall@IoU:0.5-0.95 21.5 #7 of 7 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

1x1 ConvolutionAbsolute Position EncodingsAdamAttentionBPEConvolutionDense ConnectionsDilated ConvolutionDropoutFSAFLabel SmoothingLayer NormalizationLinear LayerMask R-CNNMulti-Head AttentionPosition-Wise Feed-Forward LayerRPNResidual ConnectionRoIAlignSoft-NMSSoftmaxStochastic DepthSwin TransformerTransformerTridentNetTridentNet Block

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections