Summary
This week saw multiple new competition benchmarks that extend image restoration evaluation beyond single-degradation settings. NTIRE 2026 and related challenges introduced standardized testbeds for day/night raindrop removal, dynamic multi-exposure fusion, adverse-weather video restoration, and in-the-wild short-form UGC video restoration, collectively pushing evaluation toward more realistic compound conditions.
Situation
Representative work highlights a persistent evaluation gap in image restoration: most established methods are developed and tested against one degradation type at a time, even though real-world images frequently contain overlapping or unpredictable corruptions. In adverse-weather restoration, existing datasets mainly cover single degradations or a narrow set of two-weather combinations, leaving complex hybrid scenarios under-represented and practical assessment difficult.
The representative papers respond by treating broader benchmarks and validation protocols as necessary infrastructure. One introduces a large paired dataset spanning 31 combinations of five weather types for hybrid-condition testing; another validates burst restoration across super-resolution, denoising, and low-light enhancement on both synthetic and real RAW data; and a third stresses evaluation across an expanding set of all-in-one restoration tasks as models grow from three to five degradation types.
Infographic (English)
Progress
NTIRE 2026 The Second Challenge on Day and Night Raindrop Removal for Dual-Focused Images: Methods and Results <See Details on Fugu-MT>
This NTIRE 2026 challenge establishes a benchmark for raindrop removal under both daytime and nighttime dual-focused conditions. It adds systematic variation in illumination and focus that prior weather-removal benchmarks did not cover.
NTIRE 2026 The 3rd Restore Any Image Model (RAIM) Challenge: Multi-Exposure Image Fusion in Dynamic Scenes (Track 2) <See Details on Fugu-MT>
This RAIM challenge introduces a benchmark for multi-exposure image fusion in dynamic scenes with motion, lighting variation, and camera jitter. It targets a compound real-world degradation setting that goes beyond the static single-degradation evaluations common in earlier work.
LoViF 2026 The First Challenge on Weather Removal in Videos <See Details on Fugu-MT>
This LoViF 2026 challenge provides a dedicated benchmark and dataset for removing adverse weather from video. It extends restoration evaluation from still images to temporal weather degradation in video sequences.
NTIRE 2026 Challenge on Short-form UGC Video Restoration in the Wild with Generative Models: Datasets, Methods and Results <See Details on Fugu-MT>
This NTIRE 2026 challenge benchmarks short-form UGC video restoration under complex real-world degradations using generative models. It broadens evaluation toward in-the-wild video content with mixed, non-synthetic corruptions.
Outlook
The week's challenge reports point toward continued expansion of restoration benchmarks from mixed image degradations into harder compound settings that combine illumination shifts, motion, focus variation, and temporal effects. This trajectory aligns with representative papers that argue for broader task coverage, hybrid-condition testing, and validation on both synthetic and real data.
A likely next direction is not simply adding more tasks but making benchmarks better match practical deployment—incorporating rarer weather types, stronger nighttime simulation, video-oriented protocols, and evaluation settings that reduce dependence on large-scale paired supervision. Future-work signals from the representative papers also point toward smoother expansion to new tasks and content-aware mechanisms, suggesting that benchmarking will increasingly emphasize scalability and robustness as unified models move across image, burst, and video restoration.
Infographic (English)
References
- Gated Multi-Resolution Transfer Network for Burst Restoration and Enhancement - Authors: Nancy Mehta, Akshay Dudhane, Subrahmanyam Murala, Syed Waqas Zamir, Salman Khan, Fahad Shahbaz Khan / <See Details on Fugu-MT> / License: CC-BY-4.0
- Restoring Images Captured in Arbitrary Hybrid Adverse Weather Conditions in One Go - Authors: Ye-Cong Wan, Ming-Wen Shao, Yuan-Shuo Cheng, Yue-Xian Liu, Zhi-Yuan Bao, De-Yu Meng / <See Details on Fugu-MT> / License: CC-BY-4.0
- Cat-AIR: Content and Task-Aware All-in-One Image Restoration - Authors: Jiachen Jiang, Tianyu Ding, Ke Zhang, Jinxin Zhou, Tianyi Chen, Ilya Zharkov, Zhihui Zhu, Luming Liang, / <See Details on Fugu-MT> / License: CC-BY-4.0