Overview:
Our project aimed to improve the quality of videos by leveraging state-of-the-art image enhancement techniques. We utilized two powerful models: Real-ESRGAN for upscaling the video resolution and an Img2Img diffusion model to generate missing content within the videos.
Methods and Techniques Used:
Real-ESRGAN: Initially, we employed Real-ESRGAN, a cutting-edge neural network, for video upscaling. This model effectively increased the resolution of the video frames, enhancing the overall visual quality.
Img2Img Diffusion Model: To address missing or blank areas within the video frames, we integrated an image-to-image CompVis/stable-diffusion-v1-4 diffusion model. This technique fills in the missed content in the videos while preserving the contextual information.
Access and Availability
Accessible via HuggingFace, Real-ESRGAN offers a suite of models tailored for super-resolution imaging and CompVis/stable-diffusion-v1-4 diffusion model to Generate photo-realistic images.
Upscaling Resolution:
Lowscaled Video Input:
Upscaled Video Output:
Kommentare