A filtered subset of the Recap-DataComp-1B dataset containing 106,230,157 rows of content classified as food or drink. The data is enriched with structured food/drink extraction from the FoodExtract-v2 model, with 91.0% of rows receiving a food/drink label. The dataset was created by mrdbourke and last updated on March 20, —.
Use Cases
- Train image classification models based on the food/drink labels.
- Develop or evaluate food-specific captioning models based on the enriched re_caption data.
- Filter large-scale multimodal datasets for food-related content based on the Stage 5 and FoodExtract labels.
- Analyze the performance of automated food extraction models like FoodExtract-v2.
Strengths
- Large scale with over 106 million rows of data.
- High proportion (91.0%) of rows are pre-classified as food/drink content.
- Enriched with structured extraction from a dedicated model (FoodExtract-v2) for 74.9% of rows.
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
- huggingface
- Collection Method
- Filtered subset of the Recap-DataComp-1B dataset, enriched with FoodExtract-v2 model outputs.
- Time Range
- null
- Freshness
- Last updated 2026-03-20 23:14:08; freshness should be verified.
- Geography
- null