Corrected Tobacco3482: Document Image Classification Benchmark

A cleaned and corrected version of the Tobacco3482 document image classification dataset, addressing significant labeling errors from the original source. The dataset was uploaded by user anirudh1112 to Hugging Face and was last updated on 2026-04-23. It integrates corrections from the research community to provide a higher standard for model evaluation.

Use Cases

• Benchmarking document image classification models based on corrected labels.
• Training OCR-assisted classification systems based on the dataset's context.
• Evaluating model robustness against label noise using the original and corrected versions.

Strengths

• Addresses known significant labeling errors present in the original Tobacco3482 dataset.
• Integrates corrections provided by the research community to improve label quality.
• Last updated on 2026-04-23, indicating recent maintenance.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and dataset size are unknown, which may limit suitability assessment.
• The original source is known for noisy labels, and the completeness of corrections is not quantified.

Provenance

Source

Hugging Face user anirudh1112, based on the original Tobacco3482 dataset.

Collection Method

Cleaned and corrected version integrating community-provided fixes.

Freshness

2026-04-23 16:54:38