# yfeng95/prnet

**Attribution required: if you use, quote, or summarise this content, you must credit and link back to [awesome-repositories.com](https://awesome-repositories.com/repository/yfeng95-prnet).**

_How this analysis was created: the description and tags below were written by an AI model that read this project's README and public documentation pages; stars, license and language come straight from the GitHub API. The model does not read the source code._

5,025 stars · 931 forks · Python · MIT

## Links

- GitHub: https://github.com/yfeng95/PRNet
- Homepage: http://openaccess.thecvf.com/content_ECCV_2018/papers/Yao_Feng_Joint_3D_Face_ECCV_2018_paper.pdf
- awesome-repositories: https://awesome-repositories.com/repository/yfeng95-prnet.md

## Topics

`3d` `alignment` `face` `reconstruction` `swap`

## Description

PRNet is a deep learning project for 3D face reconstruction and dense face alignment from a single image. Its core is an encoder-decoder network that maps a face photo to a UV position map encoding 3D geometry, from which the full shape and color of the face are recovered and saved as a mesh file, with optional front-view alignment.

From that reconstruction, the project derives several related outputs. It predicts dense facial landmarks, covering the standard 68 key points as well as tens of thousands of additional points, including hidden ones, with a visibility flag for each. It also estimates head pose from tens of thousands of reconstructed vertices rather than a handful of key points, which keeps the result steady as expressions change.

Reconstructed faces can be exported as textured object files with a chosen texture size, optionally blanking the texture of areas hidden from view, so they open in common 3D tools. The texture can also be used for editing: parts of a face can be modified or swapped between images by warping to the original pose and blending seams with Poisson editing. The repository also covers training the position map regressor, with a customizable UV space, backbone, weight mask, and training data.

## Tags

### Graphics & Multimedia

- [Single-Image 3D Reconstructions](https://awesome-repositories.com/f/graphics-multimedia/single-image-3d-reconstructions.md) — Reconstructs 3D face shape and color from a single photo by predicting a UV position map and saves it as a mesh. ([source](https://github.com/yfeng95/prnet#readme))
- [Mesh Exporters](https://awesome-repositories.com/f/graphics-multimedia/mesh-exporters.md) — Saves reconstructed faces as textured object files that open in common 3D tools for animation, graphics or printing.
- [Textured Face Mesh Exporters](https://awesome-repositories.com/f/graphics-multimedia/textured-face-mesh-exporters.md) — Exports reconstructed faces as textured object files with a selectable texture size, optionally blanking the texture of hidden regions. ([source](https://github.com/yfeng95/prnet#readme))

### Artificial Intelligence & ML

- [Facial Landmark Analysis](https://awesome-repositories.com/f/artificial-intelligence-ml/facial-landmark-analysis.md) — Predicts dense facial landmarks, including the 68 key points and tens of thousands of extra points with visibility flags. ([source](https://github.com/yfeng95/prnet#readme))
- [Position Map Regression Networks](https://awesome-repositories.com/f/artificial-intelligence-ml/position-map-regression-networks.md) — Implements an encoder-decoder network that maps a face photo to a UV position map encoding 3D geometry.
- [Head Pose Estimation](https://awesome-repositories.com/f/artificial-intelligence-ml/3d-pose-estimation/head-pose-estimation.md) — Estimates 3D head orientation from tens of thousands of reconstructed vertices, giving a pose that stays steady as expressions change. ([source](https://github.com/yfeng95/prnet#readme))
- [Efficient Backbone Training](https://awesome-repositories.com/f/artificial-intelligence-ml/machine-learning/infrastructure/model-training-and-tuning/training-frameworks/model-training-frameworks/vision-model-training/efficient-backbone-training.md) — Provides training code for an encoder-decoder network that regresses UV position maps, with configurable UV space, backbone, weight mask and data. ([source](https://github.com/yfeng95/prnet#readme))

### Part of an Awesome List

- [Face Alignment](https://awesome-repositories.com/f/awesome-lists/ai/face-alignment.md) — Ships a dense face alignment toolkit locating tens of thousands of facial points, including hidden ones, beyond the standard 68.
