This is the final project for Deep Learning and NLP Research Methodologies course, NTHU. We propose our debluring model to handle OCR blurry input. Hence, we adopt the SmartDocQA, a real-world phone took text images dataset, as our deblur dataset. After our aligment of dataset and training, finally we get better OCR accuracy. report : https://drive.google.com/file/d/1YN7MRtP-H9nG6fFzkbsAzOiep2lWzLnj/view?usp=sharing