Removing Multiple Hybrid Adverse Weather in Video via a Unified Model

Ye-Cong Wan, Ming-Wen Shao Yuan-Shuo Cheng, Jun Shu, Shui-Gen Wang · 2025

Read Removing Multiple Hybrid Adverse Weather in Video via a Unified Model by Ye-Cong Wan, Ming-Wen Shao Yuan-Shuo Cheng, Jun Shu, Shui-Gen Wang (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-06200.txt.

Description

Yecong Wan and four colleagues build one model that strips every kind of bad weather out of video at once. Real storms mix rain, snow, haze, and more in shifting blends, while older tools train a separate model per weather type and fail on mixtures. Paired training videos of mixed weather barely existed either.

Their UniWRV model prompts its spatial features with learned weather priors chosen per clip. A routing module then picks how to fuse frames over time, instance by instance. To train it, the team synthesized a data set of fifteen hybrid conditions across fifteen hundred paired clips. Real storm footage tests the generalization claim.

Catalog record

CREATORYe-Cong Wan, Ming-Wen Shao Yuan-Shuo Cheng, Jun Shu, Shui-Gen Wang
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE61 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-06200

Browse the full library · About the Tower of Babel