LoDEOT: Low-Dimensional and Efficient Offset Tokens for Building Footprint Extraction from Off-Nadir Imagery

2026-10-05

AI Summary

This study addresses the inefficiency of high-dimensional instance tokens in predicting roof-to-footprint offsets for building footprint extraction from oblique imagery. We propose a compact five-dimensional offset token grounded in a local pinhole projection assumption, pioneering the replacement of high-dimensional instance tokens with a structured prior constructed through intrinsic shape and relative geometry factorization. This representation is integrated with a concentration-gated evidential mapping scheme, independent two-dimensional readout heads, and a denoising query alignment mechanism to enable efficient end-to-end prediction. Evaluated on the BONAI dataset, the proposed method achieves an FAP50 of 54.58 and a mean endpoint error (mEPE) of 5.23 pixels, outperforming the baseline by 7.56 to 16.85 percentage points. These results validate both the effectiveness and accuracy advantages of our approach.

Abstract

Instance-level roof-to-footprint offset (RFO) prediction is central to extracting building footprints from off-nadir imagery. Query-based pipelines commonly use high-dimensional instance tokens to predict signed two-dimensional RFOs. We investigate whether RFO prediction can instead use a compact offset token. Under local pinhole projection and vertical-extrusion assumptions, the idealized RFO map admits a five-parameter sufficient descriptor comprising intrinsic shape, composite amplitude, and relative geometry. This factorization provides a structural prior for a five-dimensional offset token, whose channels learn task-relevant latent representations through end-to-end training. Based on this design, we propose LoDEOT, which retains high-dimensional instance tokens for detection and segmentation but maps instance-token, concentration-gated roof, and box-mask evidence to a five-dimensional offset token followed by an independent two-dimensional readout. Known denoising-query target indices further align each supervised decoder-layer estimate with the same clean instance RFO, organizing successive predictions as target-aligned recovery under perturbed query conditions. Experiments on five real-world building datasets demonstrate the effectiveness of LoDEOT for building footprint extraction. Experiments on real-world building datasets demonstrate that a five-dimensional offset token can support accurate RFO prediction. On BONAI, LoDEOT achieves the best roof-detection bAP and bAP50 and leads all five offset-corrected footprint metrics among the evaluated end-to-end methods, with FAP50 of 54.58 and mEPE of 5.23 pixels. Its FAP50 exceeds those of the evaluated end-to-end baselines by 7.56-16.85 percentage points.

Problem

• building footprint extraction
• off-nadir imagery
• roof-to-footprint offset
• instance-level prediction
• offset token

Innovation

• Low-dimensional offset tokens
• Roof-to-footprint offset prediction
• Building footprint extraction
• Off-nadir imagery
• Denoising-query alignment