Imgscr Model Jun 2026

# Run inference with torch.no_grad(): sr_tensor = model(lr_tensor)

Researchers now use synthetic data from text-to-image models (Stable Diffusion, ControlNet) to generate infinite, perfectly labeled scene text examples—reducing the need for expensive manual annotation. imgscr model