Meta Self-Learning: Multi-Source Domain Adaptation Benchmark for OCR

This benchmark for multi-source domain adaptation in OCR and text recognition was developed by bupt-ai-cz for the ICCV 2021 conference. It provides a framework and data for evaluating meta self-learning techniques across diverse visual domains. The repository serves as a standardized evaluation point for models attempting to generalize text recognition across varying source distributions.

Use Cases

• Evaluating multi-source domain adaptation algorithms using OCR text labels
• Training meta-learning models for text recognition across different visual styles
• Benchmarking self-learning techniques for computer vision tasks

Strengths

• ICCV 2021 peer-reviewed benchmark status
• Specialized for multi-source domain adaptation
• Focuses on OCR and text recognition tasks

Limitations

• Specific record counts and file sizes are not documented in the metadata
• Highly specialized for domain adaptation research rather than general-purpose OCR training

Provenance

Source

bupt-ai-cz (Beijing University of Posts and Telecommunications), ICCV 2021

Collection Method

Compiled as a benchmark for research publication

Freshness

Last updated April 2025; based on 2021 research publication.