Deep Video Deinterlacing

Problem

Restore high-quality progressive frames from interlaced video while preserving motion details and reducing combing artifacts in a real-time setting.

Method

This project implements the approach from the paper “Real-time Deep Video Deinterlacing” using TensorFlow, with a deep network designed to reconstruct missing scanlines from interlaced inputs efficiently.

My Role

I worked on the deep video deinterlacing pipeline and released an implementation with pretrained models, example inputs, and demo results.

Links

• GitHub Repository
• Paper

Focus

• Real-time reconstruction quality
• Reducing interlacing artifacts
• Practical TensorFlow implementation for reproducible results