Imgae matting

Here are a few effects(omitting mask images)：

 

How to Run

Method 1: Run from Source Code

Firstly, you need to download the project code and install the required dependencies.

# Python 3.10

git clone https://github.com/ihmily/image-matting.git
cd image-matting
pip install -r requirements.txt

Next, use the following command to run the web interface.

python app.py

Finally, visit http://127.0.0.1:8000/.

 

Method 2: Run with Docker

Simply run the following commands after entering the project folder.

Pull the Docker image.

docker pull ihmily/image-matting:latest

After the image is pulled, run the container.

docker run -p 8000:8000 ihmily/image-matting:latest

Alternatively, you can build the image yourself.

docker build -t image-matting:latest .
docker run -p 8000:8000 image-matting:latest

Lastly, visit http://127.0.0.1:8000 to perform online image matting. Feel free to choose the method that suits your preference.

 

Use API

Please run it before use API

File upload

import requests

server = "http://127.0.0.1:8000"
image_path = "image.png"
model_name = "universal"  # people,universal
files = {"image": (image_path, open(image_path, "rb"))}
data = {"model": model_name}
response = requests.post(server+'/matting', files=files, data=data)
print(response.text)
json_data = response.json()
image_url = json_data['result_image_url']
mask_url = json_data['mask_image_url']
print("image_url:", server + image_url)
print("mask_url:", server + mask_url)

Url upload

import requests

server = "http://127.0.0.1:8000"
image_url = "http://your-image-url/demo.png"
data = {"image_url": image_url, "model": "universal"}  # people,universal
response = requests.post(server+'/matting/url', json=data)
print(response.text)
json_data = response.json()
image_url = json_data['result_image_url']
mask_url = json_data['mask_image_url']
print("image_url:",server+image_url)
print("mask_url:",server+mask_url)

You can freely choose the method you want to upload from above.If you want to get the cropped cutout, you can call crop_image_by_alpha_channel function.

 

Extended Gallery

 

References

https://modelscope.cn/models/damo/cv_unet_universal-matting/summary

https://modelscope.cn/models/damo/cv_unet_image-matting/summary