Restoring Visibility through Rain Accumulation and Veiling Effects

This direction is split into other directions: atmospheric veiling effect. Its last version stays readable and citable here.

Version 1 of 164 papers84 conceptsfrozen 2026-10-04

In computer vision, the problem of "deraining" was traditionally framed as the removal of discrete, sparse, and linear streaks from an image. However, researchers like Wenhan Yang and Robby T. Tan observed that in heavy precipitation, the visual degradation is not merely additive noise. Instead, it manifests as a dense, fog-like "rain veil" that obscures distant objects and reduces global contrast. This phenomenon, known as rain accumulation, represents a shift from simple signal separation to complex atmospheric restoration.

The Physics of the Rain Veil

The core intuition behind recent advances is that heavy rain is a multi-component degradation. While nearby raindrops appear as distinct streaks, distant droplets are too small and numerous to be resolved individually. Instead, they scatter light, creating a volumetric scattering effect similar to haze or fog.

Traditional models that treat rain as a simple additive layer—$I = B + S$, where $I$ is the rainy image, $B$ is the background, and $S$ is the streak layer—fail in torrential conditions. As noted in Deep Joint Rain Detection and Removal from a Single Image, effective restoration requires a hybrid model that accounts for both local occlusion (streaks) and global atmospheric scattering (accumulation). This "veiling effect" is structurally different from streaks; it is a global, low-frequency degradation that desaturates colors and blurs edges.

Depth Dependency and Atmospheric Scattering

A critical insight in this research direction is that rain accumulation is depth-dependent. Just as in dehazing, the further an object is from the camera, the more rain droplets the light must pass through, and the denser the "veil" becomes.

Xiaowei Hu and colleagues addressed this in Depth-Attentional Features for Single-Image Rain Removal by introducing DAF-Net. This architecture uses scene depth to guide an attention mechanism, allowing the model to apply stronger restoration to distant, heavily veiled regions while focusing on streak removal in the foreground. Similarly, Wending Yan proposed a Self-Aligned Video Deraining method that enforces physics-based consistency between scene depth and rain transmission, ensuring that the restored visibility aligns with the physical reality of light transport.

Bridging the Sim-to-Real Gap

For years, deraining models were trained on synthetic datasets where streaks were simply "Photoshopped" onto clean images. These models often failed in the real world because they lacked the "rain veil." Researchers have recently moved toward physically grounded data collection to capture these complex interactions.

  • Real-World Paired Data: Yunhao Ba and Howard Zhang developed GT-RAIN, a large-scale dataset of real-world rainy and clean image pairs. This allowed models to learn the actual appearance of rain accumulation rather than a mathematical approximation.
  • Light Transport Automation: In WeatherStream, the authors used outdoor webcams to identify "magic moments" where weather changed but the scene remained static, capturing the transition from clear skies to heavy, misty rain.
  • 3D Rendering: To handle nighttime conditions, Fan Zhang utilized the Grand Theft Auto V engine in GTAV-NightRain to model how rain streaks and accumulation refract and reflect complex urban light sources.

Joint Detection and Recursive Refinement

Because rain accumulation is so dense, removing it in a single pass often leads to artifacts or over-smoothing. Modern architectures like PReNet, introduced by Dongwei Ren and Wangmeng Zuo, use Progressive Image Deraining Networks to strip away rain layers stage-by-stage.

This recursive approach is often paired with multi-task learning. For instance, the JORDER framework jointly detects where the rain is (creating a binary mask) and removes it. By identifying the regions most affected by accumulation, the network can adaptively apply dehazing-like operations to restore contrast without "washing out" the textures in clearer parts of the image. This transition from "streak subtraction" to "joint restoration" allows models to handle the most severe weather conditions that previously baffled standard algorithms.

Papers in this version

Showing 60 of 64 papers.

History & citationVersion 1 · 2026-10-04

Split into other directions · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_restoring_visibility_through_rain_accumu_41d7729b,
  title = {Restoring Visibility through Rain Accumulation and Veiling Effects},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/restoring-visibility-through-rain-accumulation-and-veiling-effects/rdr_41d7729bdd764836a23d4a2bb407a8b0},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_restoring_visibility_through_rain_accumu_41d7729b_v1,
  title = {Restoring Visibility through Rain Accumulation and Veiling Effects},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/restoring-visibility-through-rain-accumulation-and-veiling-effects/rdr_41d7729bdd764836a23d4a2bb407a8b0/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}