Human Segmentation Tool

Upload an image, select a model and a task. Two segmentation pipelines are available — each with different strengths depending on your image and use case.

Note: Running on CPU — inference takes 3-15 seconds per image depending on model choice and image size.

Select Model

Two-stage detect-then-segment pipeline. YOLO26 locates each person with a bounding box, then UNET segments the body within each crop. Fast and instance-aware — handles multiple people separately. Fine-tuned on a combined dataset of LIP, COCO-person, MADS, and Penn-Fudan.

Select Task

Original image with detected person highlighted in green. Useful for inspecting model output quality.

Example Images

Model: YOLO26s (detection) + UNET ResNet-50 (segmentation)
Training data: LIP · COCO-person · MADS · Penn-Fudan
GitHub: Human Segmentation Evaluation