🙂Moving-Emoticon-Generation🤥

Tobigticon - Create Your Own Moving Emoticons Based on Image2Video.

Tobigticon is a service that provides the production of your own moving emoticon using GAN based image2video method.

Our model, learned by unsupervised manner, generates an emotional video of an image without ground truth through network blending parmas and landmark generation.

EmoGE'T, which is derived from Emoticon GEnerated by Tobigs, is a team of 8 members of Tobigs who worked on an image2video based project.

Make your own moving emoticons with Tobigticons!
Following the below options, the model creates your own animated emoticon.

Select an Image Style

Animation	Baby	Painting

Select an Emotion

Happiness	Disgusted	Surprised

DEMO SCREENSHOT

You can see more information about web demo here.

Requirements

We have tested on:

• CUDA 11.0
• python 3.8.5
• pytorch 1.7.1
• numpy 1.19.2
• opencv-python 4.5.1
• dlib 19.21.1
• scikit-learn 0.24.0
• Pillow 8.1.0
• Ninja 1.10.0
• glob2 0.7

Usage

Generate your own Emoticon

You can generate your own moving emoticon :)

python emoticon_generate.py --file ImagePath --transform Animation --emotion Emotion --type OutputType

For example,

python emoticon_generate.py --file 00001.jpg --transform baby --emotion disgusted --type mp4 

Training

Training the landmark generation model using the Sol1 approach.

python sol1/train.py --data_path DataPath --conditions Conditions

Train the video generation model using sol2

# Train the model
python sol2/train.py --image_discriminator PatchImageDiscriminator --video_discriminator CategoricalVideoDiscriminator --dim_z_category 3 --video_length 16  
# Generate the video using the model
python sol2/generate_videos.py [model path] [image] [class] [save_path]

Pretrained Checkpoints

Animation Baby Painting

Samples

Animation - Happy

Blend style	Generate video

Baby - Happy

Blend style	Generate video

Painting - Happy

Blend style	Generate video

Animation - Surprise

Blend style	Generate video

Baby - Surprise

Blend style	Generate video

Reference

• Rosinality, stylegan2-pytorch, 2019, https://github.com/rosinality/stylegan2-pytorch
• PieraRiccio, stylegan2-pytorch, 2019, https://github.com/PieraRiccio/stylegan2-pytorch
• justinpinkney, toonify, 2020, https://github.com/justinpinkney/toonify
• marsbroshok, face-replace, 2016, https://github.com/marsbroshok/face-replace
• sergeytulyakov, mocogan, 2017, https://github.com/sergeytulyakov/mocogan
• Yaohui Wang, Piotr Bilinski, Francois Bremond, Antitza Dantcheva. ImaGINator: Conditional Spatio-Temporal GAN for Video Generation. 2019.

Contributor 🌟

MinJung Shin

YeJi Lee

YuMin Lee

Hyebin Choi

MinKyeong Kim

SangHyeon Kim

JaeYoon Jeong

YuJin Han