PixCon

Clean-Positive Contrastive Learning for Foundation-Model Semi-Supervised Segmentation

📄 arXiv:2607.03068 Pascal VOC Cityscapes ADE20K

DINOv2-Base encoder. Upload an image and pick a label set, or click an example below. The first run per model warms up (~1 min on CPU); after that it is fast.

Label set / model

Overlay opacity

0 1

Input

Segmentation overlay

Detected classes

Run a segmentation to see detected classes.

Examples (click to run)

Input	Label set / model	Overlay opacity

PixCon · single DINOv2-Base backbone · no inference-time parameters · arXiv:2607.03068