NTIRE 2025 Challenge on Short-form UGC Video Quality Assessment and Enhancement: KwaiSR Dataset and Study

TL;DR

The paper tackles the lack of realistic benchmarks for short-form UGC image super-resolution by introducing KwaiSR, a two-part dataset with 1,800 synthetic image pairs and 1,900 wild images designed to reflect real degradation on Kwai and organized under the NTIRE 2025 Track 2. It surveys existing SR datasets, backbones, and diffusion-based approaches, then empirically benchmarks multiple diffusion-based methods against transformer-based baselines on both synthetic and wild subsets, highlighting strong perceptual gains but limited generalization to wild data. Key findings reveal that diffusion-based SR methods can achieve superior perceptual quality, yet current objective metrics often fail to align with human judgments for short-form UGC, underscoring a need for better QA metrics and more robust modeling. The dataset and challenge thus provide a practical, real-world benchmark to spur advances in short-form UGC SR and call for advances in evaluation methodology and efficient diffusion-based restoration.

Abstract

In this work, we build the first benchmark dataset for short-form UGC Image Super-resolution in the wild, termed KwaiSR, intending to advance the research on developing image super-resolution algorithms for short-form UGC platforms. This dataset is collected from the Kwai Platform, which is composed of two parts, i.e., synthetic and wild parts. Among them, the synthetic dataset, including 1,900 image pairs, is produced by simulating the degradation following the distribution of real-world low-quality short-form UGC images, aiming to provide the ground truth for training and objective comparison in the validation/testing. The wild dataset contains low-quality images collected directly from the Kwai Platform, which are filtered using the quality assessment method KVQ from the Kwai Platform. As a result, the KwaiSR dataset contains 1800 synthetic image pairs and 1900 wild images, which are divided into training, validation, and testing parts with a ratio of 8:1:1. Based on the KwaiSR dataset, we organize the NTIRE 2025 challenge on a second short-form UGC Video quality assessment and enhancement, which attracts lots of researchers to develop the algorithm for it. The results of this competition have revealed that our KwaiSR dataset is pretty challenging for existing Image SR methods, which is expected to lead to a new direction in the image super-resolution field. The dataset can be found from https://lixinustc.github.io/NTIRE2025-KVQE-KwaSR-KVQ.github.io/.

Figures (7)

• Figure 1: The visualization of our proposed KwaiSR dataset.
• Figure 2: The eleven semantic categories of the KwaiSR dataset. The left/right images are LR/HR, respectively.
• Figure 3: The semantic categories and their corresponding image counts for synthetic datasets.
• Figure 4: The semantic categories and their corresponding image counts for wild datasets.
• Figure 5: The KVQ Score distribution of different semantic categories (left) and the histogram of the overall KVQ Score distribution (right) of the wild train dataset.