DiningBench: Fine-Grained Food Classification and Nutrition Estimation Benchmark

A multimodal benchmark dataset for fine-grained food classification, nutrition estimation, and visual question answering tasks. The dataset, authored by Meituan, includes JSON Lines annotations for three tasks in both Chinese and English-translated pairs, along with a corresponding archive of images. Official evaluation scripts and instructions are hosted on the associated GitHub repository.

Use Cases

• Train models for fine-grained food classification based on image data.
• Develop algorithms for nutrition estimation from visual food data.
• Benchmark visual question answering systems on food-related queries.
• Evaluate multimodal AI systems that process both images and text annotations.

Strengths

• Designed for three distinct tasks: fine-grained food classification, nutrition estimation, and visual question answering.
• Includes paired annotations in both Chinese and English-translated versions.
• Official evaluation scripts and instructions are provided in the GitHub repository.

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

Meituan

Freshness

Last updated 2026-04-16 09:44:01; freshness should be verified.