Confidence Guided Matting (CGM) adopted by BEN2 is its core technological breakthrough, and the pipeline realizes intelligent and progressive processing through a confidence assessment mechanism. The workflow is divided into two stages: first, the base network generates the initial segmentation results and corresponding confidence heatmaps, and then the refinement network performs special optimization processing for low-confidence regions (usually complex edges). This design concept ensures intelligent allocation of computational resources, which both ensures the overall processing speed and improves the accuracy in critical areas.
Specific performance improvements are reflected in:
- Hairline-level accuracy up to 90% or more accuracy
- Translucent Objects Processing Effect Enhancement 40%
- Edge jag reduction up to 60%
Compared to traditional one-stage processing methods, the CGM technology improved overall segmentation quality scores by 35 percentage points in Prama LLC's internal tests, especially in areas such as medical image analysis and fashion e-commerce, where edge accuracy is critical.
This answer comes from the articleBEN2: Deep learning model for fast background removal from images, videosThe
