Rembg is a tool to remove images background. That is it.

• python 3.8 or newer

• torch and torchvision stable version (https://pytorch.org)

Go to https://pytorch.org and scrool down to INSTALL PYTORCH section and follow the instructions.

For example:

PyTorch Build: Stable (1.7.1)
Your OS: Windows
Package: Pip
Language: Python
CUDA: None

The install cmd is:

pip install torch==1.7.1+cpu torchvision==0.8.2+cpu -f https://download.pytorch.org/whl/torch_stable.html

Install it from pypi

pip install rembg

Remove the background from a remote image

curl -s http://input.png | rembg > output.png

Remove the background from a local file

rembg -o path/to/output.png path/to/input.png

Remove the background from all images in a folder

rembg -p path/to/input path/to/output

Copy the custom-model.pth file to ~/.u2net and run:

curl -s http://input.png | rembg -m custom-model > output.png

Start the server

rembg-server

Open your browser to

http://localhost:5000?url=http://image.png

Also you can send the file as a FormData (multipart/form-data):

<form action="http://localhost:5000" method="post" enctype="multipart/form-data">
   <input type="file" name="file"/>
   <input type="submit" value="upload"/>
</form>

In app.py

import sys
from rembg.bg import remove

sys.stdout.buffer.write(remove(sys.stdin.buffer.read()))

Then run

cat input.png | python app.py > out.png

In app.py

from rembg.bg import remove
import numpy as np
import io
from PIL import Image

input_path = 'input.png'
output_path = 'out.png'

# Uncomment the following line if working with trucated image formats (ex. JPEG / JPG)
# ImageFile.LOAD_TRUNCATED_IMAGES = True

f = np.fromfile(input_path)
result = remove(f)
img = Image.open(io.BytesIO(result)).convert("RGBA")
img.save(output_path)

Then run

python app.py

Just run

curl -s http://input.png | docker run -i -v ~/.u2net:/root/.u2net danielgatis/rembg:latest > output.png

Sometimes it is possible to achieve better results by turning on alpha matting. Example:

curl -s http://input.png | rembg -a -ae 15 > output.png

Original	Without alpha matting	With alpha matting (-a -ae 15)

• https://arxiv.org/pdf/2005.09007.pdf
• https://github.com/NathanUA/U-2-Net
• https://github.com/pymatting/pymatting

Liked some of my work? Buy me a coffee (or more likely a beer)

Copyright (c) 2020-present Daniel Gatis

Licensed under MIT License