FeyNoBg
State-of-the-art model for automatic background removal
FeyNoBg is a state-of-the-art automatic background removal model developed by Feyn Labs. It achieves top performance across eight benchmarks, matching or beating leaders on four metrics. The model is built on an enhanced BiRefNet architecture with an expanded third stage (24 blocks instead of 18) to better balance foreground recognition and boundary precision.
Background removal is challenging for complex images with hair, motion blur, or low contrast. FeyNoBg handles these by combining localization and reconstruction modules. The model was trained on a diverse dataset of 26.1K images from 10 sources, covering crowded scenes, camouflage, high resolution, portraits, and anime.
FeyNoBg is open source and available through the NoBg library on GitHub or as a pre-trained model on Hugging Face. It is designed for developers, researchers, and anyone needing high-quality background removal.
Key Features
State-of-the-Art Performance
FeyNoBg posts the best published S-measure on four of eight benchmarks and comes within 2% of the leader on the rest, including ultra-high-resolution datasets like UHRSD-TE.
