UGround V1 Data: Web-Hybrid Images with Bounding Box Annotations

UGround V1 Data is a computer vision dataset from the osunlp organization, last updated on May -1, 2025. The description notes an addition of a bounding box version of 'Web-Hybrid' data, with normalized coordinates. Access to the dataset requires an application process involving contact via email.

Use Cases

• Train object detection models based on the bounding box annotations mentioned in the description.
• Benchmark model performance on web-hybrid imagery likely containing diverse objects.
• Preprocess image data for model input based on the normalized coordinate system described.

Strengths

• Includes a bounding box version with normalized coordinates (0,999), as stated in the update.
• Last updated on 2025-05-02, indicating recent maintenance.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Access is gated by an approval process, which may delay usage.

Provenance

Source

osunlp organization on Hugging Face.

Collection Method

Likely contains web-hybrid imagery, suggesting a mix of web-crawled and synthetic sources.

Time Range

null

Freshness

Last updated 2025.05.02 07:04:40

Geography

null