A DCGAN to generate anime faces using custom mined dataset

Pavitrakumar P

Last update: Jan 3, 2023

Related tags

Deep Learning python deep-learning anime keras cnn generative-adversarial-network gan face

Overview

Anime-Face-GAN-Keras

A DCGAN to generate anime faces using custom dataset in Keras.

Dataset

The dataset is created by crawling anime database websites using curl. The script anime_dataset_gen.py crawls and processes the images into 64x64 PNG images with only the faces cropped.

Examples of the dataset:

Network

This implementation of GAN uses deconv layers in Keras (networks are initialized in the GAN_Nets.py file). I have tried various combinations of layers such as :
Conv + Upsampling
Conv + bilinear
Conv + Subpixel Upscaling
But none of these combinations yielded any decent results. The case was either GAN fails to generate images that resembles faces or it generates same or very similar looking faces for all batches (generator collapse). But these were my results, maybe techniques such as mini-batch discrimination, z-layers could be used to get better results.

Training

Only simple GAN training methods are used. Training is done on about 22,000 images. Images are not loaded entirely into memory instead, each time a batch is sampled, only the sampled images are loaded. An overview of what happens each step is:
-Sample images from dataset (real data)
-Generate images using generator (gaussian noise as input) (fake data)
-Add noise to labels of real and fake data
-Train discriminator on real data -Train discriminator on fake data
-Train GAN on fake images and real data labels
Training is done for 10,000 steps. In my setup (GTX 660; i5 4670) it takes 10-11 secs for each step.

Loss plot:

Full Training as a GIF: (images sampled every 100 step)

Faces generated at the end of 10,000 steps:

The faces look pretty good IMO, might look more like an actual face with more training, more data and probably with a better network.

Resources

https://github.com/tdrussell/IllustrationGAN
https://github.com/jayleicn/animeGAN
https://github.com/forcecore/Keras-GAN-Animeface-Character

https://distill.pub/2016/deconv-checkerboard/
https://kivantium.net/keras-bilinear

You might also like...

A simple rest api serving a deep learning model that classifies human gender based on their faces. (vgg16 transfare learning)

this is a simple rest api serving a deep learning model that classifies human gender based on their faces. (vgg16 transfare learning)

5 Dec 9, 2021

High accurate tool for automatic faces detection with landmarks

faces_detanator High accurate tool for automatic faces detection with landmarks. The library is based on public detectors with high accuracy (TinaFace

7 May 10, 2022

Code of 3D Shape Variational Autoencoder Latent Disentanglement via Mini-Batch Feature Swapping for Bodies and Faces

3D Shape Variational Autoencoder Latent Disentanglement via Mini-Batch Feature Swapping for Bodies and Faces Installation After cloning the repo open

37 Dec 3, 2022

Example-custom-ml-block-keras - Custom Keras ML block example for Edge Impulse

Custom Keras ML block example for Edge Impulse This repository is an example on

8 Nov 2, 2022

PyTorch implementation of a Real-ESRGAN model trained on custom dataset

Real-ESRGAN PyTorch implementation of a Real-ESRGAN model trained on custom dataset. This model shows better results on faces compared to the original

160 Jan 4, 2023

A repo to show how to use custom dataset to train s2anet, and change backbone to resnext101

3 Dec 28, 2022

For holding anime-related object classification and detection models

Animesion An end-to-end framework for anime-related object classification, detection, segmentation, and other models. Update: 01/22/2020. Due to time-

72 Nov 30, 2022

[Open Source]. The improved version of AnimeGAN. Landscape photos/videos to anime

4.4k Dec 27, 2022

StyleGAN2 Webtoon / Anime Style Toonify

StyleGAN2 Webtoon / Anime Style Toonify Korea Webtoon or Japanese Anime Character Stylegan2 base high Quality 1024x1024 / 512x512 Generate and Transfe

121 Dec 21, 2022

Comments

Assistance editing GAN for different image set

Hi @pavitrakumar78,

I wanted to get in touch to see if you could possibly assist with an issue I am having. I am using your GAN_training.py to see if it can be applied to a different image set. The dimensions of the image are 3232x208. The images themselves are of video game level. As such I tried to change your code in the GAN_training.py so that it could use the game levels images as inputs. I also wish to output the generated images from the GAN as single images and not as 4x4 images as your code does.

Please see screenshots attached that indicate areas where I have changed your code. Screenshot 1 - I have changed the figsize and gridspec size to 1x1

screenshot2 - changed the image shape to 3232x208, the shape of the images i am trying to use as well as batch size and number of steps.

screenshot3 - changed the image_shape in the GAN_Nets.py file for get_disc_normal function

screenshot4 - error message i receive when running the GAN_training.py

I would really appreciate your help, or any guidance you can give me as to how to resolve the issue shown in screenshot 4. Please let me know if you need anymore information.

Thanks.

opened by vinaysp 7
Program Installation files missing

Hello, i have this issue where i download the folder, but in the folder there's no .exe file to install the application. Could you help me find the .exe file?

opened by Blazeolmo 1
Generating the refined image

Hey Pavitra,

Really loved your implementation of Anime Face GAN. I just had one minor doubt. I am new in the field of ML. So I was wondering that since the dataset consists of around 1000 images, our output generated in the jupyter notebook codes shows only 16 images (maybe just a glimpse of some images). But how can I generate/recreate the rest of the images? Like do I need to upload the image with a new python code in a fresh jupyter notebook and then again train my model on that single image? I would really appreciate it if you can help me out with this.

opened by psychedelic2007 1