jantic/DeOldifyArchived
DeOldify
DeOldify is a deep learning system and a set of pre-trained computer vision models designed to apply realistic colors to grayscale photographs and video footage. It functions as a neural media restoration tool that uses trained networks to estimate original hues for black-and-white media and remove glitches and artifacts from aged images and film.
The project employs a NoGAN colorization technique that removes the GAN discriminator during training to prevent artifacts and avoid over-saturation of pixels. For cinematic sequences, it applies temporal frame consistency to maintain color stability and prevent flickering between consecutive frames.
The framework covers broad capability areas including deep learning image processing and media restoration. It utilizes generative adversarial networks, convolutional neural architectures, and automatic color labeling to synthesize realistic color values based on learned patterns from large datasets.
Features
- General Image Colorization - Provides deep learning models to apply realistic colors to black-and-white photographs.
- Computer Vision Models - Ships pre-trained computer vision models designed to estimate original hues for black-and-white media.
- Colorization Tools - Provides a tool to transform black-and-white video into color while preventing frame flickering.
- Convolutional Neural Networks - Employs deep convolutional neural networks to map grayscale intensities to color coordinates.
- Deep Learning Image Processors - Automates the visual restoration and colorization of legacy media using neural networks.
- Generative Adversarial Image Synthesis - Uses generative adversarial networks to synthesize realistic colors based on learned patterns from large datasets.
- Temporal Stability Constraints - Applies temporal stability constraints to prevent color flickering across consecutive video frames.
- Media Restoration Tools - Removes glitches and artifacts from aged images and film to produce high-fidelity restoration.
- AI Video Colorization - Applies realistic color restoration to grayscale video sequences using temporal consistency.
- Automatic Color Labeling - Implements automatic color labeling to generate ground truth data for training colorization models.
- Model Weights - Utilizes frozen weights from pre-trained convolutional networks for high-level feature extraction.
- NoGAN Colorization - Utilizes a NoGAN technique that removes the discriminator during training to eliminate color artifacts.
- Portrait Colorization Models - Uses deep learning models to estimate and apply original hues to monochrome images.
- Computer Vision - Deep learning based image and video colorization.
- Computer Vision Libraries - Deep learning project for restoring and colorizing old images.
- Image and Video Restoration - Tool for colorizing and restoring old photos and videos.
- Image Generation And Editing - Deep learning model for colorizing and restoring old media.
- Style Transfer - Listed in the “Style Transfer” section of the The Incredible Pytorch awesome list.
Star history
How this analysis was created: This summary and feature list were written by an AI model that read the project's README and public documentation pages. Each feature links to the documentation it came from; stars, license and language come straight from the GitHub API. The model does not read the source code, and the analysis is refreshed when the project is re-analysed. Learn more on our About page.
