Image Matting Example

Introduction

Image	Result

Image Matting is the process of accurately estimating the foreground object in images and videos. It is a very important technique in image and video editing applications, particularly in film production for creating visual effects.

Available Models

• RMBG v1.4: Faster processing with good quality results
• RMBG v2.0: Higher quality results with increased processing time
• RMBG v2.0 Quantized: Optimized version of v2.0 with better CPU performance

Usage

1. Import images (Ctrl+I) or a video (Ctrl+O) into X-AnyLabeling.
2. Select and load the desired model version (RMBG v1.4, RMBG v2.0, or RMBG v2.0 Quantized).
3. Click Run (i) to process the current image. After checking the result, press Ctrl+B to process all images.

The output, once completed, will be automatically stored in a x-anylabeling-matting subdirectory within the same folder as your original image.

Performance Benchmark

Performance test processing 10 images with different model configurations:

Model Version	Device	Processing Time (seconds)
RMBG v1.4	GPU	2
RMBG v1.4	CPU	9
RMBG v2.0	GPU	13
RMBG v2.0	CPU	122
RMBG v2.0 Quantized	GPU	55
RMBG v2.0 Quantized	CPU	105

Test Environment

• CPU: 13th Gen Intel(R) Core(TM) i5-13600K
• GPU: NVIDIA GeForce RTX 3060 (12GB VRAM)

Note

RMBG v2.0 models are significantly slower on CPU. For optimal performance, GPU acceleration is highly recommended.