VisCoT VStar Collage: Multimodal QA Data for Visual Search Training

m-Just's dataset comprises collages with a randomly placed 'core' image and a corresponding question-answer pair. This data was used to train the vSearcher model introduced in the research paper 'InSight-o3: Empowering Multimodal Foundation Models with Generalized Visual Search'. The dataset was last updated on Hugging Face in January 2026.

Use Cases

• Train visual search models based on the described collage-and-QA structure.
• Benchmark model performance on locating and answering questions about a 'core' image within a collage.
• Study visual reasoning and attention in multimodal AI based on the described filler and core image sampling method.

Strengths

• Designed for training a specific visual search model (vSearcher) mentioned in a research paper.
• Each entry contains a structured multimodal unit (collage + QA pair) as described.
• Last update timestamp (2026-01-29) is provided.

Limitations

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

Provenance

Source

Hugging Face, author m-Just.

Collection Method

Created as part of the training data for the vSearcher model, using images from VisualCoT and V* training data.

Time Range

null

Freshness

Last updated 2026-01-29 07:00:20; freshness should be verified.

Geography

null