mazhar004 / modnet-bgremover

A deep learning approach to remove background & adding new background image

MODNet Background Remover

Application

A deep learning approach to remove background and adding new background image

• Remove background from images,videos & live webcam
• Adding new background to those images,videos & webcam footage

Demo

Before removing the background	After replacing the background with new image
Before removing the background from video	After replacing the background with new image in this video

Web View

Before removing the background	After removing the background

Installation

Python Version

• Python == 3.8

Virtual Environment

Windows

• python -m venv venv
• .\venv\Scripts\activate
• If any problem for scripts activation
  • Execute following command in administration mode
    • Set-ExecutionPolicy Unrestricted -Force
  • Later you can revert the change
    • Set-ExecutionPolicy restricted -Force

Linux

• python -m venv venv
• source venv/bin/activate

Library Installation

• Library Install
  • pip install --upgrade pip
  • pip install --upgrade setuptools
  • pip install -r requirements.txt
  • To run in web interface
    • pip install -r web_requirements.txt

Pretrained Weights Download

• Weights Detail

Inference

Image

Single image

It will generate the output file in output/ folder

• python inference.py --image image_path [Without background image]
• python inference.py --image image_path --background True [With background image]
• Example:
  • python inference.py --image assets/sample_image/female.jpeg
  • python inference.py --image assets/sample_image/male.jpeg --background True

Folder of images

It will generate the output file in output/ folder

• python inference.py --folder folder_path [Without background image]
• python inference.py --folder folder_path --background True [With background image]
• Example:
  • python inference.py --folder assets/sample_image/
  • python inference.py --folder assets/sample_image/ --background True

Video

It will generate the output file in output/ folder

• python inference.py --video video_path [Without background image]
• python inference.py --video video_path --background True [With background image]
• Example:
  • python inference.py --video assets/sample_video/sample.mp4
  • python inference.py --video assets/sample_video/sample.mp4 --background True

Webcam

• python inference.py --webcam True [Without background image]
• python inference.py --webcam True --background True [With background image]

Webinterface

• python api.py
• Click on this link/localhost
• Upload the image and wait

Reference

• A Trimap-Free Solution for Portrait Matting in Real Time under Changing Scenes
• Sample Female photo by Michael Dam on Unsplash
• Sample Male photo by Erik Lucatero on Unsplash

modnet-bgremover's Issues

Request: passing webcam to video device.

Thank you for the good work. It gives excellent results when cropping heads, even with curly hair.

But unfortunately the result cannot be used as a device with the webcam option. It would be nice if such an output would be added to the script.

Unfortunately, I have no idea about programming, but I imagine to forward the signal to v4linux or OBS.

Removing white Background

Hi,
I am using the code to remove a white background from the image but the results are not as expected. Any ideas on how to make it work. Attaching the result.

Resulant image

Original image

Did you train the pre-training model yourself？

The pre-training model is ZHKKKe trained？or you trained again?

Erro happend 'NoneType' object has no attribute 'shape'

while running 'python3 inference.py --image assets/sample_image/femalclee.jpeg'
got this error Erro happend 'NoneType' object has no attribute 'shape'

UPDATE: its my bad

ask about image transparent

Hello, I want to ask. how to make background image transparent? i need transparent background image png. Thank You

Quality Output Image

Hello, I want ask. I need to change resolution image because i want
use image to print. can you help me? Thank you

CUDA error: no kernel image is available for execution on the device

I've installed BGRemover to Linux. I had some problems with torchvision version, because pip did not understand requirements.txt for some reason.

I have the following packages installed

Package Version

---

numpy 1.24.1
opencv-python 4.5.1.48
Pillow 6.2.2
pip 22.3.1
setuptools 65.6.3
six 1.16.0
torch 1.7.1
torchvision 0.2.2.post3
typing_extensions 4.4.0

When I run python inference.py --image

I get "Erro happend CUDA error: no kernel image is available for execution on the device"

So, should I modify torchvision version, install some additional pip package, both or something else?

Transparent output

Can the output be the video with transparent background and no image?

White lines surrounding the images

Hello, I ran your model but i see some white lines surrounding my object. Do you have any idea how to make those white lines disappear?