What is pgtformer?

PGTFormer is a tool that fixes blurry, pixelated, or damaged faces in videos. If you have an old or low-quality video clip where someone's face looks blocky or unclear, this project can sharpen and restore that face to look more natural and detailed, all automatically. What makes it stand out is that it doesn't require "pre-alignment," a step where many similar tools need faces to be manually adjusted to a standard position before processing. This tool skips that step entirely and works on faces as they naturally appear in the video. It also focuses on keeping faces consistent from frame to frame, so the restored result looks smooth rather than flickering or shifting as the video plays. The audience for this is anyone working with video restoration, say, a filmmaker cleaning up archival footage, a media company remastering old interviews, or a researcher building better face-enhancement tools. The project includes example videos so you can test it quickly, and the creators provide pre-trained models that download automatically the first time you run it, so you don't need to train anything yourself. Benchmark results in the README show it outperforming several other well-known methods across standard quality metrics, meaning the restored faces score higher in sharpness, structural similarity, and perceptual quality. The project is built on top of existing open-source frameworks (BasicSR and CodeFormer), and it was trained using a large dataset of high-quality face videos called VFHQ. It was published as a research paper at the AI conference IJCAI in 2024. The code and instructions are aimed at someone comfortable running Python commands, but the README keeps the setup fairly straightforward if you have basic familiarity with installing Python packages.