acceptodds
Desk rejected at ICLR 2027

A Benchmark High-Resolution Multi-Temporal Remote Sensing Dataset for Urban Tree Semantic Segmentation

Desk rejected at ICLR 2027, so left out of the paper lists and the map.

Abstract

Trees are vital to urban ecosystems and the global carbon cycle. Accurate tree identification and segmentation support urban greening monitoring, ecological assessment, and carbon sink estimation. However, existing datasets mostly rely on single-temporal imagery, limiting studies of long-term tree dynamics. To address this, we construct Tree-RGB-MT, a multi-temporal urban tree segmentation dataset based on high-resolution RGB aerial imagery (2013–2022) from multiple cities and seasons, with fine-grained pixel-level annotations. Experiments with state-of-the-art semantic segmentation models demonstrate that Tree-RGB-MT poses significant challenges in boundary complexity, scene diversity, and temporal consistency. Thus, the dataset provides a robust benchmark for studying tree dynamics, phenological variation, and urban greening evolution.

Primary area:
datasets and benchmarks
Keywords:
multi-temporal urban tree dataset, high-resolution remote sensing imagery, semantic segmentation, tree segmentation

Related papers

Discussion (0)

Sign in to comment.