I develop computer-vision systems and the supporting tools needed to take them from an experiment to a reliable workflow. My interests include image processing, dataset preparation, segmentation, synthetic-data generation, and reproducible model inference.
The repositories highlighted below grew out of a larger computer-vision project. I needed a practical way to prepare and expand its datasets, so I built several reusable utilities around segmentation, compositing, and controlled image generation. I published them separately because I thought they might be useful to other engineers facing similar problems.
A reusable pipeline for creating coherent training images from foreground objects and new scene backgrounds. It combines segmentation, compositing, inpainting, and structural guidance to reduce visible seams and domain mismatch.
- Foreground extraction with BiRefNet
- Configurable compositing and inpainting-mask strategies
- Automatic CUDA, MPS, and CPU selection
- Reproducible run metadata and model-free regression tests
A set of modular tools originally built to create and manage synthetic datasets for a computer-vision project.
- Unified interfaces for segmentation, generation, and guided inpainting
- Background replacement and dataset augmentation workflows
- Explicit VRAM/RAM lifecycle management
- Indexed dataset-file management and reproducible seeds
Python · PyTorch · OpenCV · ONNX Runtime · Pillow · Diffusers · SDXL · ControlNet · uv · Git
- Computer-vision engineering and applied machine learning
- Dataset design, preparation, and augmentation
- Image segmentation, processing, and analysis
- Reliable and reproducible inference pipelines
I'm open to remote opportunities and collaborations.
