ICCV 2025

HoliTracer: Holistic Vectorization of Geographic Objects from Large-Size Remote Sensing Imagery

Wuhan University
*Corresponding author

Holistic Vectorization Tracing in 1080P: Conventional patch-based vectorization cuts large geographic objects into fragmented tiles. HoliTracer performs holistic vectorization across full-scene satellite imagery: starting from an initial contour, the active Polygon Sequence Tracer (PST) window walks along the sequence, autoregressively snapping points and lighting up golden corner vertices to synthesize a seamless, topologically closed vector polygon.

Abstract

With the increasing resolution of remote sensing imagery (RSI), large-size RSI has emerged as a vital data source for high-precision vector mapping of geographic objects. Existing methods are typically constrained to processing small image patches, which often leads to the loss of contextual information and produces fragmented vector outputs. To address these, this paper introduces HoliTracer, the first framework designed to holistically extract vectorized geographic objects from large-size RSI. In HoliTracer, we enhance segmentation of large-size RSI using the Context Attention Net (CAN), which employs a local-to-global attention mechanism to capture contextual dependencies. Furthermore, we achieve holistic vectorization through a robust pipeline that leverages the Mask Contour Reformer (MCR) to reconstruct polygons and the Polygon Sequence Tracer (PST) to trace vertices. Extensive experiments on large-size RSI datasets, including buildings, water bodies, and roads, demonstrate that HoliTracer outperforms state-of-the-art methods.

Method Overview

HoliTracer Framework Overview

The Overall Framework of HoliTracer: HoliTracer is composed of three interconnected modules: (1) Context Attention Net (CAN) for multiscale semantic segmentation of ultra-large satellite imagery; (2) Mask Contour Reformer (MCR) for robust initial polygon boundary reconstruction; and (3) Polygon Sequence Tracer (PST) for autoregressive vertex identification, geometric corner refinement, and topological contour closure. This holistic architecture eliminates tile boundary cutting and produces seamless GIS-grade vector polygons.

Quantitative Results

Method PoLiS (↓) CIoU (%) AP APs APm APl IoU (%) F1 (%)
TS-MTA 6.94 14.29 36.33 14.34 65.49 41.30 85.30 92.02
LCF-ALE 7.51 19.09 41.78 20.05 64.44 38.98 85.67 92.22
DeepSnake 4.43 53.54 38.83 22.31 53.17 15.47 76.05 86.01
E2EC 9.10 47.66 29.85 17.03 41.32 8.87 65.44 78.49
FFL 9.34 35.51 31.39 17.03 48.29 26.92 71.48 83.25
UniVec 8.36 62.26 30.13 16.86 42.18 27.90 78.66 87.96
HiSup 5.49 42.23 56.77 36.08 69.44 37.82 85.79 92.21
HoliTracer (Ours) 3.63 82.30 61.07 40.37 80.30 60.00 91.60 95.41
* Performance comparison on the WHU-building large-size RSI benchmark (10,000×10,000 pixels). HoliTracer achieves state-of-the-art vector quality with lowest PoLiS distance (3.63) and highest polygon CIoU (82.30%).

Qualitative Results

HoliTracer Qualitative Results

Qualitative Vectorization Comparison: Holistic vectorization results across diverse geographic object classes—dense residential buildings, intricate hydrological networks, and multi-lane arterial road systems. HoliTracer maintains sharp corners, regular geometry, and closed topology across full-scene imagery without tile seams.

Citation

@inproceedings{wang2025holitracer,
  title={HoliTracer: Holistic Vectorization of Geographic Objects from Large-Size Remote Sensing Imagery},
  author={Wang, Yu and Dang, Bo and Li, Wanchun and Chen, Wei and Li, Yansheng},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
  year={2025}
}