It was inspired by Auto-LaMa.
Unlike Auto-Lama, it differs in:
- Use the object instance segmentation model MaskDINO instead of the object detection model DETR.
- Use LaMa with refiner for better results.
simple demo with gradio
A minimum of 12 gb memory gpu is required.
.root
├─demo.py
├─ckpt
│ ├──maskdino_swinl_50ep_300q_hid2048_3sd1_instance_maskenhanced_mask52.3ap_box59.0ap.pth
│ └─models
│ ├──config.yaml
│ └─models
│ └─best.ckpt
└─images
├──buildings.png
├──cat.png
└──park.png
- conda environment setup
conda create --name maskdino python=3.8 -y
conda activate maskdino
conda install pytorch==1.9.0 torchvision==0.10.0 cudatoolkit=11.1 -c pytorch -c nvidia
pip install -U opencv-python
mkdir repo
git clone [email protected]:facebookresearch/detectron2.git
cd detectron2
pip install -e .
pip install git+https://github.com/cocodataset/panopticapi.git
cd ..
git clone -b quickfix/infer_demo --single-branch https://github.com/MeAmarP/MaskDINO.git
cd MaskDINO
pip install -r requirements.txt
cd maskdino/modeling/pixel_decoder/ops
python setup.py build install
cd ../../../../..
git clone https://github.com/geomagical/lama-with-refiner.git
cd lama-with-refiner
pip install -r requirements.txt
pip install --upgrade numpy==1.23.0
cd ../..
pip install gradio
- Run
#localhost http:https://127.0.0.1:7860
python demo.pyMany thanks to these excellent opensource projects