ARGUS is a framework for calculating hallucination and omission costs in free-form video captions. The dataset, created by tomg-group-umd, provides metrics to quantify the degree of hallucinated and omitted content in video-language model outputs. It was last updated on June 10,我们发现了一个问题，您提供的原始描述中包含了中文文本。根据指令，我需要将输入翻译成英文。以下是翻译后的描述，并基于此生成输出。
Use Cases
- Benchmarking video-language model performance based on hallucination and omission cost metrics.
- Training or fine-tuning models to reduce hallucinated content in video descriptions.
- Training or fine-tuning models to reduce omitted content in video descriptions.
- Analyzing the trade-offs between detail and accuracy in generated video captions.
- Comparing different Video-LLM architectures based on their ArgusCost-H and ArgusCost-O scores.
Strengths
- Framework provides two specific, defined metrics: ArgusCost-H for hallucination and ArgusCost-O for omission.
- Dataset is associated with a published paper and a dedicated website, suggesting academic rigor.
- Last updated on 2025-06-10, indicating recent maintenance.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count and dataset scale are unknown, which may limit suitability assessment.
- The specific video content and caption data used for evaluation are not described.
Provenance
- Source
- tomg-group-umd on Hugging Face.
- Collection Method
- Likely contains scores calculated by the ARGUS framework on video-caption pairs.
- Time Range
- null
- Freshness
- Last updated 2025-06-10 02:30:08; freshness should be verified.
- Geography
- null