Removes backgrounds while preserving transparent, camouflaged, typographic, and glow details.
Fine-tunes a BiRefNet image segmentation architecture for difficult alpha matting cases.
Provides MIT-licensed code and Hugging Face model weights.
Benchmarks performance over camouflage, transparency, complex scenes, hair, text, effects, illustration, and design categories.
Includes Python loading examples through transformers and torchvision preprocessing.
Offers a CLI workflow for local image background removal.
Includes a FastAPI serving path with model, refinement, and decontamination options.
Supports browser testing through a linked Hugging Face demo.