LINEAiGE
AI HISTORY

U-Net Research Publication

Olaf Ronneberger, Philipp Fischer, and Thomas Brox submitted 'U-Net: Convolutional Networks for Biomedical Image Segmentation' on May 18, 2015. The paper presented a network and training strategy for biomedical image segmentation built around a contracting path for context and a symmetric expanding path for precise localization.

18 May 2015AI HISTORYVERIFIED

Historical classification

This verified record belongs to LINEAiGE's artificial-intelligence history layer beginning in 1956.

Why it matters

A documented convolutional-network architecture milestone for biomedical image segmentation that paired a contracting path with a symmetric expanding path and demonstrated end-to-end training from comparatively few annotated images.

People, organizations & technologies

PEOPLE
Olaf Ronneberger, Philipp Fischer, Thomas Brox

TECHNOLOGIES
U-Net, Convolutional neural networks, Biomedical image segmentation

Evidence & verification

STATE
verified

CONFIDENCE
high

LAST VERIFIED
2026-09-10T10:04:24+08:00

The original arXiv record establishes the May 18, 2015 first-submission date, authors, and the documented contracting/expanding architecture and training claims. The University of Freiburg Computer Vision Group publication record independently corroborates the paper and its 2015 MICCAI publication. This record uses May 18 strictly as a research-publication/submission milestone; it does not infer a software-release date, institutional geography, ancestry, later influence, causality, or relationships.

Sources

SOURCE 1 · PRIMARY · CLAIM-EVIDENCE

U-Net: Convolutional Networks for Biomedical Image Segmentation

arXiv

PUBLISHED
2015-05-18

SOURCE 2 · PRIMARY · CORROBORATION

U-Net: Convolutional Networks for Biomedical Image Segmentation

University of Freiburg Computer Vision Group

Trace the chronology

Correction history

Correction 1

DATE
2026-09-10

PREVIOUS
institutional-record

NOTE
Normalized the University of Freiburg corroborating source type from the candidate-only label institutional-record to the canonical-compatible sourceType value other; source identity and evidence claims are unchanged.

Record identity

LINEAiGE ID
u-net-2015