Open Source Background Removal Model BEN2, Capable of Removing Backgrounds from Images and Videos

Recently, PramaLLC launched its latest artificial intelligence model - the Background Erase Network BEN2. This model has made significant innovations in foreground segmentation technology. BEN2 utilizes a Confidence Guided Matting (CGM) pipeline, employing a refined network specifically designed to handle pixels with low confidence from the base model, resulting in more precise and reliable matting effects.

BEN2's training dataset includes DIS5k and PramaLLC's proprietary 22K segmentation dataset. Thanks to these high-quality data, BEN2 excels in hair matting, 4K image processing, object segmentation, and edge optimization. Notably, the base model is open-source, allowing users to try it out by visiting its official website, or integrate BEN2 into their projects via API.

For developers, installing BEN2 is quite simple. You can complete the installation using the pip command, and then quickly start image processing tasks with just a few lines of code. The model supports single image and batch image processing, with a recommended maximum batch size of 3 to ensure optimal performance on consumer-grade GPUs.

In addition to image processing, BEN2 also offers video segmentation capabilities. Users can easily separate the foreground from the background in videos, with a straightforward operation process. Simply specify the video path, and BEN2 will automatically process it, ultimately saving the generated video as foreground.webm or foreground.mp4 for convenient later use.

PramaLLC provides a free online demo on its official website, allowing users to personally experience the powerful features of BEN2.

Model: https://huggingface.co/PramaLLC/BEN2

Online use on HuggingFace: https://huggingface.co/spaces/PramaLLC/BEN2

Official website: https://backgrounderase.net/home

Key Points:

🌟 BEN2 utilizes confidence-guided matting technology to enhance image matting accuracy.

🖼️ Supports single and batch image processing, with a user-friendly operation.

🎥 Provides video segmentation features, making it easy for users to handle video foreground and background.