Blur Anything is an adaptation of the excellent Track Anything project which is in turn based on Meta's Segment Anything and XMem. It allows you to blur anything in a video, including faces, license plates, etc.
# Clone the repository:
git clone https://github.com/Y-T-G/Blur-Anything.git
cd Blur-Anything
# Install dependencies:
pip install -r requirements.txt
# Run the Blur-Anything gradio demo.
python app.py --device [cpu|cuda:0|cuda:1|...] --sam_model_type [vit_t| vit_b|vit_h| vit_l] [--backend [onnx|openvino]]
- FastSAM with ONNX and OpenVINO support.
- Lower memory usage.
- Add a gradio demo
- Add support to use YouTube video URL
- Add option to completely black out the object
- Convert XMem to ONNX
The project is an adaptation of Track Anything which is based on Segment Anything and XMem.
Thanks to PIMS which is used to process video files while keeping memory usage low.