Dockerized fork of renatoviolin/bg-remove-augment

Remove image backgrounds using the U2-Net pre-trained model.

• Remove background → transparent PNG
• Replace background → composite with custom image
• Dataset augmentation → batch combine with multiple backgrounds
• CPU & GPU support → choose the right image for your hardware

CPU:

docker run -p 8000:8000 ghcr.io/r-oswald/bg-remove-augment:cpu

GPU (requires NVIDIA Container Toolkit):

docker run --gpus all -p 8000:8000 ghcr.io/r-oswald/bg-remove-augment:gpu

services:
  bg-remove:
    image: ghcr.io/r-oswald/bg-remove-augment:cpu  # or :gpu
    ports:
      - 8000:8000
    volumes:
      - ./input:/app/bg-remove-augment-docker/webapp/images-input
      - ./output:/app/bg-remove-augment-docker/webapp/images-output

docker build -f Dockerfile.cpu -t bg-remove:cpu .
docker build -f Dockerfile.gpu -t bg-remove:gpu .

• U2-Net: github.com/xuebinqin/U-2-Net

@article{Qin_2020_PR,
  title   = {U2-Net: Going Deeper with Nested U-Structure for Salient Object Detection},
  author  = {Qin, Xuebin and Zhang, Zichen and Huang, Chenyang and Dehghan, Masood and Zaiane, Osmar and Jagersand, Martin},
  journal = {Pattern Recognition},
  volume  = {106},
  pages   = {107404},
  year    = {2020}
}