This dataset was created by Tongji University and King's College London in the paper "AdaTreeFormer: Few shot domain adaptation for tree counting from a single high-resolution image" was proposed in the "

This paper contains three datasets: London dataset, Yosemite dataset and Jiangsu dataset.

This dataset is a high-resolution tree detection dataset in London.

• Location: Yosemite National Park, California, United States
• Landscape type: Woody mountainous area
• Average number of trees per image: 36
• Total number of trees: 98,949
• Image resolution: 0.12 m
• Data division: training set: 1350 images, test set: 1350 images

The Yosemite dataset covers primarily woody mountainous areas with low tree density and complex terrain, providing an important testing environment for the performance of models in complex terrain.

Dataset background

• Diverse tree types and terrain: Different types, sizes, and shapes of trees, as well as different terrains (e.g., urban, farmland, mountainous) make tree counting more complicated.
• Lack of high-quality training data: Deep learning models usually rely on large amounts of labeled data, but this data is expensive and time-consuming to obtain.
• Domain gap problem: In the tree counting task, different scenes (such as urban and rural), different imaging types (such as aerial images and satellite images), and different tree densities can lead to significant differences between the source domain and the target domain.