SPREAD: A large-scale, high-fidelity synthetic dataset for multiple forest vision tasks

Open Webpage

Zhengpeng Feng, Yihang She, Srinivasan Keshav

Published: 01 Jan 2025, Last Modified: 24 Oct 2025Ecol. Informatics 2025EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•Introduced SPREAD, a synthetic forest dataset with 55k samples for various vision tasks.•SPREAD pretraining reduces real data need by 75% for tree trunk segmentation task.•Properly combining SPREAD with real datasets enhances segmentation accuracy.•SPREAD is fully open-source, enabling customizable extension for specific needs.