Youku Video Enhancement and Super-Resolution Competition Announcement

Competition Overview

The Youku Video Enhancement and Super-Resolution Challenge is now open. Participants can find the competition entry page at https://tianchi.aliyun.com/competition/entrance/231711/introduction . Video enhancement and super‑resolution are core computer‑vision algorithms that aim to restore degraded video content and improve visual clarity. These technologies have significant industrial value, especially for restoring early‑film footage.

Dataset

The competition provides the largest and most diverse industry dataset for video super‑resolution and enhancement, containing over 10,000 video pairs with varied content categories, noise models, and difficulty levels. The dataset simulates real‑world business noise patterns, allowing researchers to test algorithms in realistic scenarios. The first 1,000 pairs are used for the competition; after the contest, about 2,000 pairs will be released publicly, and the remaining 7,000 will be gradually opened.

Prizes

Champion: 1 team, RMB 100,000 + certificate Runner‑up: 1 team, RMB 60,000 + certificate Third place: 1 team, RMB 40,000 + certificate Geek Award: 3 teams, RMB 10,000 each + certificate The top 10 teams that pass the semi‑final review also gain access to Alibaba (Youku) campus recruitment green channel.

Schedule

Pre‑selection: May 5 – June 18 (UTC+8) Semi‑final: June 25 – August 10 (UTC+8) Final: September (UTC+8)

Registration

Registration opens: April 11 2019 (UTC+8)

Deadline for registration and team changes: June 16 2019 12:00 (UTC+8)

Teams of 1‑5 members; each participant may join only one team.

Accurate registration information is required; otherwise the team may be disqualified.

Register via the Tianchi website using a Taobao or Alibaba Cloud account.

Competition Task

Participants must train models that convert low‑resolution video frames (width/height ratio 1:4) into high‑resolution frames. The high‑resolution videos come from Youku’s HD media library, while the low‑resolution videos are generated by simulating real‑world noise and compression artifacts.

Task Analysis

The challenge goes beyond pure super‑resolution; it also requires handling noise, compression artifacts, and other quality degradations. Existing resources such as the “A collection of high‑impact and state‑of‑the‑art SR methods” repository (https://github.com/thstkdgus35/EDSR-PyTorch) and NTIRE competition reports provide useful baselines. Notable methods include:

EDSR – the winner of NTIRE 2017, which removes unnecessary residual modules and introduces a multi‑scale model.

Meta‑SR – a recent approach that supports arbitrary (including non‑integer) scaling factors via a Meta‑Upscale module that predicts filter weights dynamically.

Future research directions highlighted by the organizers include achieving fast, high‑quality upscaling while simultaneously mitigating noise and compression artifacts, which is a key challenge for real‑world video applications.

Reference Papers and Resources

Super‑resolution: Google’s “RAISR: Rapid and Accurate Image Super Resolution” (https://arxiv.org/abs/1606.01299) and the NTIRE 2018 challenge report (http://www.vision.ee.ethz.ch/~timofter/publications/NTIRE2018_SR_report_CVPRW-2018.pdf). Video denoising and compression‑artifact removal: “Multi‑level Wavelet‑CNN for Image Restoration”, “Residual Dense Network for Image Restoration”, and “Image Restoration Using Convolutional Auto‑encoders with Symmetric Skip Connections”. Video enhancement: NTIRE enhancement winners such as Meitu’s “Range Scale Global Unet” (2018) and “Modified HRNet” (2019).

The competition aims to bridge industrial needs and academic research, fostering deeper collaboration between the two communities.

We wish all participants great success in the challenge.