akanimax/T2F akanimax/T2F

  • Stars
    star
    546
  • Rank 80,822 (Top 2 %)
  • Language
    Python
  • License
    MIT License
  • Created about 6 years ago
  • Updated over 2 years ago
Twitter logo Share LinkedIn logo Share Facebook logo Share
akanimax/T2F
github

akanimax

Reviews

There are no reviews yet. Be the first to send feedback to the community and the maintainers!

Repository Details

T2F: text to face generation using Deep Learning

! Attention !

This project is unfortunately not being worked upon

Please head over to the following much cooler project that takes the idea of Text-2-Image generation to the next level:

DallE: Original PyTorch

โญ [NEW] โญ

T2F - 2.0 Teaser (coming soon ...)

2.0 Teaser

Please note that all the faces in the above samples are generated ones. The T2F 2.0 will be using MSG-GAN for the image generation module instead of ProGAN. Please refer link for more info about MSG-GAN. This update to the repository will be comeing soon ๐Ÿ‘.

T2F

Text-to-Face generation using Deep Learning. This project combines two of the recent architectures StackGAN and ProGAN for synthesizing faces from textual descriptions.
The project uses Face2Text dataset which contains 400 facial images and textual captions for each of them. The data can be obtained by contacting either the RIVAL group or the authors of the aforementioned paper.

Some Examples:

Examples

Architecture:

Architecture Diagram

The textual description is encoded into a summary vector using an LSTM network. The summary vector, i.e. Embedding (psy_t) as shown in the diagram is passed through the Conditioning Augmentation block (a single linear layer) to obtain the textual part of the latent vector (uses VAE like reparameterization technique) for the GAN as input. The second part of the latent vector is random gaussian noise. The latent vector so produced is fed to the generator part of the GAN, while the embedding is fed to the final layer of the discriminator for conditional distribution matching. The training of the GAN progresses exactly as mentioned in the ProGAN paper; i.e. layer by layer at increasing spatial resolutions. The new layer is introduced using the fade-in technique to avoid destroying previous learning.

Running the code:

The code is present in the implementation/ subdirectory. The implementation is done using the PyTorch framework. So, for running this code, please install PyTorch version 0.4.0 before continuing.

Code organization:
configs: contains the configuration files for training the network. (You can use any one, or create your own)
data_processing: package containing data processing and loading modules
networks: package contains network implementation
processed_annotations: directory stores output of running process_text_annotations.py script
process_text_annotations.py: processes the captions and stores output in processed_annotations/ directory. (no need to run this script; the pickle file is included in the repo.)
train_network.py: script for running the training the network

Sample configuration:

# All paths to different required data objects
images_dir: "../data/LFW/lfw"
processed_text_file: "processed_annotations/processed_text.pkl"
log_dir: "training_runs/11/losses/"
sample_dir: "training_runs/11/generated_samples/"
save_dir: "training_runs/11/saved_models/"

# Hyperparameters for the Model
captions_length: 100
img_dims:
  - 64
  - 64

# LSTM hyperparameters
embedding_size: 128
hidden_size: 256
num_layers: 3  # number of LSTM cells in the encoder network

# Conditioning Augmentation hyperparameters
ca_out_size: 178

# Pro GAN hyperparameters
depth: 5
latent_size: 256
learning_rate: 0.001
beta_1: 0
beta_2: 0
eps: 0.00000001
drift: 0.001
n_critic: 1

# Training hyperparameters:
epochs:
  - 160
  - 80
  - 40
  - 20
  - 10

# % of epochs for fading in the new layer
fade_in_percentage:
  - 85
  - 85
  - 85
  - 85
  - 85

batch_sizes:
  - 16
  - 16
  - 16
  - 16
  - 16

num_workers: 3
feedback_factor: 7  # number of logs generated per epoch
checkpoint_factor: 2  # save the models after these many epochs
use_matching_aware_discriminator: True  # use the matching aware discriminator

Use the requirements.txt to install all the dependencies for the project.

$ workon [your virtual environment]
$ pip install -r requirements.txt

Sample run:

$ mkdir training_runs
$ mkdir training_runs/generated_samples training_runs/losses training_runs/saved_models
$ train_network.py --config=configs/11.comf

Other links:

blog: https://medium.com/@animeshsk3/t2f-text-to-face-generation-using-deep-learning-b3b6ba5a5a93
training_time_lapse video: https://www.youtube.com/watch?v=NO_l87rPDb8
ProGAN package (Seperate library): https://github.com/akanimax/pro_gan_pytorch

#TODO:

1.) Create a simple demo.py for running inference on the trained models

More Repositories

1

BMSG-GAN

[MSG-GAN] Any body can GAN! Highly stable and robust architecture. Requires little to no hyperparameter tuning. Pytorch Implementation
Python
630
star
2

pro_gan_pytorch

Unofficial PyTorch implementation of the paper titled "Progressive growing of GANs for improved Quality, Stability, and Variation"
Python
536
star
3

msg-stylegan-tf

MSG StyleGAN in tensorflow
Python
264
star
4

natural-language-summary-generation-from-structured-data

Implementation of the paper -> https://arxiv.org/abs/1709.00155. For converting information present in the form of structured data into natural language text
Python
183
star
5

Variational_Discriminator_Bottleneck

Implementation (with some experimentation) of the paper titled "VARIATIONAL DISCRIMINATOR BOTTLENECK: IMPROVING IMITATION LEARNING, INVERSE RL, AND GANS BY CONSTRAINING INFORMATION FLOW" (arxiv -> https://arxiv.org/pdf/1810.00821.pdf)
Python
152
star
6

msg-gan-v1

MSG-GAN: Multi-Scale Gradients GAN (Architecture inspired from ProGAN but doesn't use layer-wise growing)
Python
151
star
7

fagan

A variant of the Self Attention GAN named: FAGAN (Full Attention GAN)
Python
112
star
8

thr3ed_atom

ReLU Fields The Little Non-linearity That Could
Python
111
star
9

big-discriminator-batch-spoofing-gan

BMSG-GAN with more features
Python
45
star
10

pro_gan_pytorch-examples

Examples trained using the python pytorch package pro-gan-pth
Python
38
star
11

attn_gan_pytorch

python package for self-attention gan implemented as extension of PyTorch nn.Module. paper -> https://arxiv.org/abs/1805.08318
Python
19
star
12

my-deity

I worship the one true neural network architecture that can autonomously learn everything.
Jupyter Notebook
12
star
13

NLP2SQL

A research and review of techniques to provide a natural language interface to RDMS.
Jupyter Notebook
11
star
14

open-styleganv2-pytorch

Open source + Free for Commercial Use implementation of StyleGANv2 in pytorch
Python
7
star
15

capsule-network-TensorFlow

The impending concept of capsule networks has finally arrived at arXiv. link to the publication -> https://arxiv.org/abs/1710.09829 . In this repository, I'll create an implementation using TensorFlow from scratch as an exercise.
Jupyter Notebook
6
star
16

3inGAN

Python
5
star
17

GAN-understanding

Implements gans on toy datasets and preliminary ML datasets for showing certain aspects of convergence and stability. Tries to cover various loss functions defined over the years.
Jupyter Notebook
5
star
18

autoencoder-cifar-10

Implementing an auto-encoder for the cifar10 dataset
Jupyter Notebook
4
star
19

Homecoming

repository for mini-projects
Python
3
star
20

python_ai_project_template

A lightweight template for building AI-based prototype/research POCs in Python. My poison (DL framework :laugh: ) of choice is PyTorch!
Python
3
star
21

some-randon-gan-1

MSG-GAN with self attention. For MSG-GAN head to -> https://github.com/akanimax/MSG-GAN
Python
2
star
22

AI-Literature

A repository to store key research works from the past. It is also an attempt to structure and organize these research papers.
2
star
23

indian-celeb-gans

Various GANs trained on a dataset containing images of Indian Celebrities (procured by me).
Python
2
star
24

some-random-gan-2

More experimentation with the base MSG-GAN architecture. This includes the coord-conv layers in the architecture. For more info about MSG-GAN, head to -> https://github.com/akanimax/msg-stylegan-tf
Python
2
star
25

multithreaded-histogram-equalization-cpp

Explanatory Code for performing Histogram Equalization on Images for contrast improvement. The code uses OpenCV in C++ for image read/write and uses pthread for multithreading
C++
2
star
26

deep-reinforcement-learning

Project for studying and implementing the traditional RL algorithms and also the DL variants of the same.
Jupyter Notebook
1
star
27

dcgan_pytorch

GAN example created using the attn_gan_pytorch package -> https://github.com/akanimax/attn_gan_pytorch
Python
1
star
28

CL-3_lab_2017

repository for assignments of Computer Laboratory 3 - 2016
TeX
1
star
29

SVC2004-deep-learning

A deep learning based solution for the SVC2004 problem.
Jupyter Notebook
1
star
30

toxic-comment-identification-tensorflow

Data -> https://www.kaggle.com/c/jigsaw-toxic-comment-classification-challenge/data
Python
1
star
31

REST_MOVIE_TICKET_SYSTEM

A restful system implementing token based authentication for allowing users to book movie tickets online. Use of Play framework for scala
Scala
1
star
32

algorithms

A repository for collecting the coding implementations of some of the most famous algorithms
Python
1
star
33

energy-preserving-neural-network

When a data signal propagates through the Neural Network, it is not mandatory that the energy of the signal will be preserved throughout the neural computations. This research attempts at collecting (perhaps creating) techniques for preserving the Energy throughout the network.
Python
1
star