nst-telegram-bot

A telegram bot providing neural style transfer service.

Try it

https://t.me/YetAnotherOneNSTBot

Setup

Docker

git clone https://github.com/masim05/nst-telegram-bot.git
cd nst-telegram-bot
docker build -t nst-telegram-bot .
docker run -d -e TG_BOT_TOKEN=<YOUR_TOKEN> -e IMAGE_SIZE=256 -e EPOCHS=500 -e LR=0.005 -e ALPHA=10  -e BETA=60 nst-telegram-bot

Native, tested with python 3.11.4

git clone https://github.com/masim05/nst-telegram-bot.git
cd nst-telegram-bot
python3 -m venv .venv
source .venv/bin/activate
pip3 install -r requirements.txt
IMAGE_SIZE=64 EPOCHS=500 LR=0.005 ALPHA=10 BETA=60 TG_BOT_TOKEN=<YOUR_TOKEN> python app.py

Supported environment variables:

• IMAGE_SIZE - size of the result image
• EPOCHS - number of epochs to learn
• LR - learning rate
• ALPHA and BETA - weights of content_loss and style_loss respectively

Usage

Available commands:

• /start - general greeting
• /help - usage info
• /debug - debug info

The bot expects to receive two images and sends result of NST in return. The first image will be used as a content image, the second - as a style one.

At the moment only one NST requests can be processes at a time due to hardware limitations of the hosting server. However, during NST other commands (eg, /debug) work.

Results

With parameters: IMAGE_SIZE=256, EPOCHS=500, LR=0.005, ALPHA=10.0, BETA=60.0

TODO

• Share the code on github
• Prepare README file
• Implement the main flow
• Dockerize
• Deploy and run
• Implement in non-blocking manner
• Implement /transfer_style command
• Re-train models

Credits

• https://docs.python-telegram-bot.org/en/stable/examples.echobot.html
• https://towardsdatascience.com/implementing-neural-style-transfer-using-pytorch-fd8d43fb7bfa
• https://arxiv.org/pdf/1508.06576.pdf