DarkGeekMS / portrait-synthesis-from-speech

This repository contains the main code for portrait synthesis from speech experiments associated with Retratista application. Refer to the application repository at https://github.com/DarkGeekMS/Retratista

DarkGeekMS/portrait-synthesis-from-speech Issues

Complete the graduation project book and final presentation
Closed 3 years ago
Complete the required project videos
Closed 3 years ago
Implement multiple head poses generation pipeline
Closed 3 years ago
Complete final speech recognition training
Closed 3 years ago
Re-train BERT multi-label text classifier for facial attributes
Closed 3 years ago
Decide speech recognition workflow
Closed 3 years ago
Re-structure the repository after completion
Closed 3 years ago
Explore alternative methods to BERT for text processing
Closed 3 years ago
Try StyleGAN2 fine-tuning for better feature separation
Closed 3 years ago
Setup normalizing flows framework for conditioned latent space manipulation
Closed 3 years ago
Expand the current set of feature directions through various method
Closed 3 years ago
Decide multiple head poses generation workflow
Closed 3 years ago1
Perform final integration and sanity check of directions-based face generation
Closed 3 years ago
Adapt StyleGAN2 face generation workflow to face refinement module
Closed 3 years ago
Transfer StyleGAN2 to PyTorch in directions-based face generation
Closed 3 years ago
Explore possible improvements for end-to-end text-to-face generation network
Closed 3 years ago1
Decide face refinement workflow
Closed 3 years ago1
Text2Face dataset processing and expansion
Closed 3 years ago
Explore pretrained embedding to latent vector mapping
Closed 3 years ago1
Re-calculate feature components of latent vector after each latent navigation
Closed 3 years ago1
Generate face from text with centralized pose and neutral facial expression
Closed 3 years ago
Centralize skewed feature directions
Closed 3 years ago1
Integrate generated feature directions to current text-to-face generation
Closed 3 years ago2
Train a multi-label classification network for facial attributes
Closed 3 years ago
Apply orthogonalization to generated feature directions
Closed 3 years ago2
Define a set of quantitative metrics to assess the system performance
Closed 3 years ago
Add debugging structure to directions-based text-to-face generation pipeline
Closed 3 years ago
Handle difference in magnitudes of generated feature directions
Closed 3 years ago
Define a complete set of facial features to work on for text-to-face generation
Closed 3 years ago
Disentangle generated feature directions
Closed 3 years ago1
Add nonlinear weighting of differentiated logits to directions-based face generation
Closed 3 years ago1
Complete directions-based staged face generation codebase
Closed 4 years ago
Implement text-to-face generation via attribute disentanglement
Closed 4 years ago1
Load the complete end-to-end text-to-face generation network for training
Closed 4 years ago1
Replace NLP module with more complicated approaches
Closed 4 years ago1
Refactor the end-to-end network codebase to accept generic sentence embedding models
Closed 4 years ago
Write a comprehensive review about latent space navigation in StyleGAN2
Closed 4 years ago1
Try open-source StyleGAN2 projectors
Closed 4 years ago1
Complete a test script for the whole network
Closed 4 years ago
Use more robust loss for reconstructed images
Closed 4 years ago1
Create a mapping network from StyleGAN2 latent space to FaceNet encoding
Closed 4 years ago1
Improve training GPU utilization
Closed 4 years ago
Replace MSE loss for latent vectors with other losses
Closed 4 years ago1
Add an option for loading pre-trained weights before training
Closed 4 years ago