trocr-small-handwritten

Model Overview

TrOCR-small-handwritten is a transformer-based optical character recognition (OCR) model fine-tuned on the IAM handwriting dataset. It uses an encoder-decoder architecture: an image Transformer (initialized from DeiT) encodes input images, and a text Transformer (initialized from UniLM) autoregressively generates text. Images are processed as sequences of 16x16 fixed-size patches with absolute position embeddings.

Key Strengths

• Designed specifically for handwritten text recognition from single text-line images.
• Leverages pre-trained vision and language transformers for robust performance on handwriting.
• Fine-tuned on the IAM dataset, a standard benchmark for handwriting recognition.

Best For

Optical character recognition (OCR) on handwritten text-line images, such as scanned documents, notes, or historical manuscripts.

Benchmark Results

The model card does not include specific benchmark numbers or evaluation metrics. The model was introduced in the paper TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models (Li et al., 2021).

Example Input

The model accepts single text-line images. A sample image from the IAM database is shown below:

Citation

Li et al., 2021: TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models