TT100K: 100,000 Street View Images with 30,000 Annotated Traffic Signs

Over 30,000 annotated traffic sign instances across 221 categories are contained in this dataset built from 100,000 Tencent Street View images. The Tsinghua–Tencent 100K (TT100K) dataset is designed for real-world detection and classification tasks, featuring variation in lighting, weather, viewpoint, and distance. It was uploaded by Genius-Society to Hugging Face, with a last recorded update in May 2026.

Use Cases

• Training traffic sign detection models based on the 30,000 annotated instances.
• Benchmarking classification algorithms across the 221 sign categories.
• Developing robust vision systems based on the described variation in lighting, weather, and viewpoint.
• Studying model performance on real-world street view imagery from the 100,000 source images.

Strengths

• Built from 100,000 source images, providing a large-scale visual corpus.
• Contains over 30,000 annotated instances, offering substantial training data.
• Covers 221 traffic sign categories, supporting fine-grained classification tasks.
• Designed with real-world variation in conditions like lighting and weather.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The description references a 45-class subset used in an original study, suggesting potential complexity in using the full 221 classes.

Provenance

Source

Tsinghua–Tencent collaboration, built from Tencent Street View images.

Collection Method

Likely gathered via automated or semi-automated annotation of street view imagery.

Freshness

Last updated 2026-05-03 08:10:00; freshness should be verified.

Geography

Likely China, based on the involvement of Tencent and Tsinghua University.