StreetTree is a benchmark dataset for fine-grained street tree classification containing over 12 million images of more than 8,300 common species. Collected from urban streetscapes across 133 countries spanning five continents, it includes expert-verified observational data and a hierarchical taxonomy. The dataset was created by Jiapeng Li and last updated in May 2026.
Use Cases
- Benchmarking vision models for fine-grained classification based on high inter-species visual similarity.
- Studying long-tailed natural distributions of urban tree species based on the dataset's global collection.
- Developing models robust to intra-class variations caused by seasonal changes, as described.
- Training classifiers to handle diverse imaging conditions like lighting, occlusions, and camera angles.
- Researching hierarchical classification and representation learning based on the provided order–family–genus–species taxonomy.
Strengths
- Contains over 12 million images, providing a large-scale resource.
- Covers more than 8,300 common street tree species.
- Geographically diverse, with data collected from urban streetscapes across 133 countries spanning five continents.
- Includes expert-verified observational data and a hierarchical taxonomy.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
- Data may reflect geographic or collection bias inherent to the sourcing method.
Provenance
- Source
- figshare
- Collection Method
- Images collected from urban streetscapes across 133 countries, supplemented with expert-verified observational data.
- Freshness
- Last updated 2026-05-05 12:47:53; freshness should be verified.
- Geography
- 133 countries spanning five continents.