NTIRE Workshop & Challenges @ CVPR 2026 • Denver, Colorado
Real-World Face Restoration Challenge
Restore high-quality (HQ) face images from real-world degraded low-quality (LQ) inputs. LQ images are often affected by blur, noise, compression, and other distortions, making this task highly ill-posed. Achieving high perceptual quality and stable generalization remains challenging.
Challenge Overview
This challenge is part of the NTIRE 2026 Workshop & Challenges co-located with CVPR 2026 in Denver, Colorado. NTIRE (New Trends in Image Restoration and Enhancement) is one of the most influential international competitions and workshops in the field of image restoration and enhancement, aiming to bring together the latest advances from academia and industry, and promote open, fair, and reproducible algorithm comparison and exchange.
Why face restoration is challenging
Face Restoration aims to restore high-quality (HQ) face images from real-world degraded low-quality (LQ) inputs. LQ images are often affected by blur, noise, compression, and other distortions, making this task highly ill-posed. In recent years, CNNs, Transformers, and generative models (such as diffusion models) have significantly improved restoration results, but achieving high perceptual quality and stable generalization under complex real-world degradations remains challenging.
Application value
The importance of face restoration is not only reflected in perceptual improvement, but also directly affects downstream tasks, including: face recognition, keypoint detection, 3D face reconstruction, etc. It is widely used in security and forensics, video surveillance, mobile/wearable devices, social media content enhancement, and other application scenarios, thus requiring higher demands for "high-quality, real-time, and efficient" methods.
Data & Competition Phases
Where to get the data
All datasets are hosted on the competition platform under the Files section.
Development: Training
Only high-quality (HQ) images are provided. Participants are encouraged to design their own degradation model to synthesize LQ inputs for training.
Development: Validation
Validation data provides only LQ inputs (for debugging and comparison).
Test Phase
Final test data provides only inputs for final evaluation. Participants submit restoration results and provide reproducible code or an executable.
Fairness requirement: Please do not use validation and test sets for training.
Evaluation
Metrics
Quality scores will be obtained by weighted averaging of all restoration results on the validation/test sets. The official description uses multiple no-reference (No-Reference) quality metrics for comprehensive evaluation (examples include: NIQE, CLIP-IQA, ManIQA, MUSIQ, FID, Q-Align, etc.; specific metric list and weights are subject to the competition page).
Submission requirements
- Submit restoration results for the test set by the result deadline.
- Submit a Fact Sheet and reproducible code/executable by the code deadline.
- Top-ranked teams will be invited to present at the NTIRE workshop.
Important Dates
Awards & Publication
Awards
Top-ranked participants will receive awards and be invited to write method descriptions according to CVPR Workshop paper submission requirements, to be submitted to the NTIRE Workshop proceedings co-located with CVPR 2026.
Proceedings
The overall results will be published in the NTIRE 2026 Workshop Proceedings (CVPR Workshops).
Join the Challenge
Competition portal
Register, download data, and submit results on the CodaBench competition page.
Open CodaBenchNTIRE 2026
Learn more about the NTIRE 2026 workshop and other challenges on the official NTIRE website.
Open NTIRE 2026Organizers
Organizer list may be updated.
WeChat Group
Scan the QR code below to join the official WeChat group for announcements and discussion.
Note: If the QR code expires, please scan the QR code at the bottom of the competition homepage to join the group.
