What is scene segmentation?

Scene segmentation (also called semantic segmentation) divides an image into regions where every pixel is assigned a class label — sky, road, building, person, tree, water, and so on. Unlike object detection (which draws bounding boxes), segmentation precisely outlines each region at the pixel level.

What AI model powers this tool?

SegFormer-B0 fine-tuned on the ADE20K dataset (150 semantic categories), published by NVIDIA and provided via Hugging Face as Xenova/segformer-b0-finetuned-ade-512-512. It runs entirely in your browser via @huggingface/transformers (transformers.js) — your images are never uploaded to a server.

Which region categories are supported?

The ADE20K dataset covers 150 semantic categories including outdoor scenes (sky, mountain, water, field, tree, road, sidewalk, building, fence) and indoor scenes (wall, ceiling, floor, furniture, window, door, person, car, and many more).

AI Scene Segmenter — Label Image Regions Free

Automatically segment any photo into labeled regions (sky, floor, wall, people, cars, trees…) using SegFormer AI running entirely in your browser. Free, no upload, no signup.

How it works

1Upload any photo by dragging it in or clicking to browse. Supports JPG, PNG, and WebP up to 20 MB.
2On first use the SegFormer-B0 model (~25 MB) downloads and caches permanently in your browser — subsequent runs are instant.
3Click "Segment Scene" — SegFormer analyses the photo and identifies every region (sky, road, building, vegetation, person, etc.) in seconds.
4The result shows a colour-coded overlay drawn on top of your original image, with a legend below listing every identified region and its confidence score.
5Click "Download Segmented PNG" to save the annotated image to your device.