Based on recent advancements in 2D facial landmark detection and work on projection of 2D landmarks into a 3D model, we propose a two step architecture for 3D facial alignment through a landmark detection approach. Our experiments show that our method outperforms a direct 3D model. This repo contains an implementation of our method.
- Direct 3D Model
- 2D model with 3D projection
Modify the config.json as needed and run:
python train.py -c config.json
You can test trained model by running inference.py passing path to the trained checkpoint by --resume argument.
Example:
python inference.py -r path/to/trained_checkpoint
You can resume from a previously saved checkpoint by:
python train.py --resume path/to/checkpoint
HandwritingGeneration/
│
├── README.md
│
├── train.py - main script to start training
├── inference.py - evaluation of trained model
│
├── config.json - holds configuration for training a model
├── parse_config.py - class to handle config file and cli options
│
├── base/ - abstract base classes
│ ├── base_data_loader.py
│ ├── base_model.py
│ └── base_trainer.py
│
├── data_loader/
│ └── data_loaders.py - Class to handle the loading of the data
│
├── data/ - default directory for storing input data
│
├── model/ - models, losses, and metrics
│ ├── models.py
│ ├── metric.py
│ └── loss.py
│
├── saved/
│ ├── figures/ - saved images of output 3D face alignment
│ ├── models/ - trained models are saved here
│ └── log/ - logdir for tensorboard and logging output
│
├── trainer/ - trainers
│ └── trainer.py
│
├── logger/ - module for tensorboard visualization and logging
│ ├── visualization.py
│ ├── logger.py
│ └── logger_config.json
│
└── utils/ - utility functions
└── util.py