LlamaSeg: Image Segmentation Masks via Autoregressive Generation

LlamaSeg is a dataset for image segmentation via autoregressive mask generation. The data is provided as JSON annotation files packaged in compressed shards, following the naming convention of the SA-1B dataset from Segment Anything. It was authored by GML-FMGroup and last updated on 2026-05-27.

Use Cases

• Training autoregressive models for image segmentation based on the described mask generation framework.
• Benchmarking segmentation model performance against the SA-1B dataset structure.
• Researching scalable annotation formats for large-scale segmentation tasks.
• Developing tools for processing sharded JSON annotation files in computer vision pipelines.

Strengths

• Data is structured to follow the established SA-1B dataset naming convention, suggesting compatibility with existing tools.
• The dataset was updated on 2026-05-27, indicating recent maintenance.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file size, and specific data format details are unknown, which may limit suitability assessment.

Provenance

Source

GML-FMGroup via Hugging Face.

Collection Method

Likely derived from or related to the SA-1B (Segment Anything 1B) dataset.

Freshness

Last updated 2026-05-27 14:40:01; freshness should be verified.