Document Image Generation

Module: Generative AI | Difficulty: Advanced

Document Layout Generation

Text Rendering Pipeline

1. Generate layout
2. Render text with font/style
3. Apply document degradation

Synthetic OCR Training Data

class DocumentGenerator(nn.Module):
    def __init__(self, layout_net, text_renderer, degradation):
        super().__init__()
        self.layout = layout_net
        self.text = text_renderer
        self.degrade = degradation
    def forward(self, text_list, style):
        layout = self.layout(len(text_list), style)
        clean = self.text(text_list, layout)
        return self.degrade(clean)

Research Insight: Synthetic document images improve OCR accuracy by 15-25% on out-of-domain documents. The key is to model realistic degradations: blur, noise, skew, and background variation.