SenseNova-SI-8M: Spatial Intelligence Training Corpus

SenseNova-SI-8M contains approximately 8.16 million training samples spanning about 2.72 million unique images. It is the official full-scale training dataset for the SenseNova-SI series, used to train the SenseNova-SI-1.1-InternVL3-8B model. The dataset was created by sensenova and last updated on May 13, 2026.

Use Cases

• Train spatial intelligence models based on the dataset's rigorous taxonomy of spatial capabilities.
• Benchmark computer vision algorithms using the curated collection of ~2.72 million images.
• Conduct research on multimodal AI using the dataset's large-scale, annotated image-text pairs.
• Fine-tune foundation models for tasks requiring spatial reasoning based on the structured training corpus.

Strengths

• Approximately 8.16 million training samples provide substantial scale.
• The dataset is organized under a rigorous taxonomy of spatial capabilities.
• It serves as the canonical training corpus for spatial intelligence research.

Limitations

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

Provenance

Source

sensenova

Freshness

Last updated 2026-05-13 04:54:38; freshness should be verified.