SENTINEL: A Dataset for Mitigating Object Hallucinations in Multimodal LLMs

A dataset associated with the ICCV 2025 paper 'SENTINEL: Mitigating Object Hallucinations via Sentence-Level Early Intervention'. The dataset was created by author psp-dada and last updated on February 11, 2026. It is designed to address the problem of fabricated content in multimodal large language models.

Use Cases

• Training models to reduce visual hallucinations based on the described sentence-level intervention approach.
• Benchmarking hallucination mitigation methods based on the dataset's focus on cross-modal understanding.
• Analyzing distribution mismatches in training data for MLLMs based on the problem description.

Strengths

• Dataset is associated with a peer-reviewed ICCV 2025 conference paper.
• Last updated on February 11, 2026, 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.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

psp-dada on Hugging Face

Collection Method

Likely created for research purposes as described in the associated ICCV 2025 paper.

Time Range

null

Freshness

Last updated 2026-02-11 18:01:11

Geography

null