StyleTransformerGANs

Final project to Deep Learning School. Style Trasformer via telegram bot.

You can check how this works in demo notebooks

About

Bot's name: @styletransferjune2020bot, try it!

UPD: It's over cause my AWS free month is over :)

When you want to impose some kind of style on an image, you can use just such a bot. The bottom line is simple: upload an image with content, upload a style, the bot itself creates a connected image. To better understand what is at stake, take a look at a few examples:

This project includes three main parts:

1. StyleTransform Network
2. Telegram Bot
3. Deploying (via AWS)

StyleTransform Network

Multi-Style-Transfer model by Hang Zhang was a key point in this project. You can check arxiv about Multi-style Generative Network for Real-time Transfer.

Telegram Bot

Bot using aiogram library for Python. To better understand how it's works, take a look at some actions:

First, you choose content image, second - style image.

Then, you need to choose quality. The lower the quality, the less time it takes to process. High quality takes about 15-30 seconds.

Bot returns you result image and type /continue command.

You can use /help command to see some help.

Deploying (via AWS)

Bot loaded using AWS, PuTTY and WinSCP. First you need to create a server using AWS. Then, using PuTTY, we start the server on Ubuntu. Using WinSCP, we run this on Windows.