sample_usage.py

# Sample code for using the Portuguese OCR Dataset

import h5py
import numpy as np
from PIL import Image
import matplotlib.pyplot as plt
from transformers import VisionEncoderDecoderModel, TrOCRProcessor
import torch

# Load the dataset
def load_dataset(file_path):
    with h5py.File(file_path, 'r') as f:
        images = f['images'][:]
        texts = [t.decode('utf-8') if isinstance(t, bytes) else t for t in f['texts'][:]]
    return images, texts

# Load train dataset
train_images, train_texts = load_dataset('train_dataset.h5')
print(f"Loaded {len(train_images)} training samples")

# Display a random sample
def display_sample(images, texts, idx=None):
    if idx is None:
        idx = np.random.randint(0, len(images))
    
    print(f"Text: {texts[idx]}")
    
    plt.figure(figsize=(12, 3))
    plt.imshow(images[idx])
    plt.axis('off')
    plt.title(f"Sample {idx}")
    plt.show()
    
    return idx

# Display a random sample
sample_idx = display_sample(train_images, train_texts)

# Example of using with TrOCR
def test_with_trocr(image, model_name="microsoft/trocr-base-printed"):
    # Load model and processor
    processor = TrOCRProcessor.from_pretrained(model_name)
    model = VisionEncoderDecoderModel.from_pretrained(model_name)
    
    # Convert image to PIL if it's a numpy array
    if isinstance(image, np.ndarray):
        image = Image.fromarray(image)
    
    # Prepare image
    pixel_values = processor(image, return_tensors="pt").pixel_values
    
    # Generate prediction
    generated_ids = model.generate(pixel_values)
    generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
    
    return generated_text

# Uncomment to test a sample with a pre-trained TrOCR model
# sample_img = train_images[sample_idx]
# predicted_text = test_with_trocr(sample_img)
# print(f"Original text: {train_texts[sample_idx]}")
# print(f"Predicted text: {predicted_text}")