Weather-Conditioned Depth Anything

European Conference on Computer Vision (ECCV) 2026

Zhaoming Xu Chan-Wei Hu Kuan-Ru Huang Zihao Zhu Renjie Li Yang Zhou Zhengzhong Tu†

Texas A&M University, College Station, TX 77843, USA

† Corresponding author

PaperComing soon CodeComing soon ModelComing soon DataComing soon

Project overview

Qualitative Depth Map Comparison
Under Adverse Weather Conditions

Drag the two handles to compare the input, Depth Anything V2, and DA-W.

✨ Abstract ✨

Monocular depth estimation foundation models, such as the Depth Anything series, have achieved remarkable performance across diverse domains. However, they still suffer from critical failures under adverse weather conditions, such as fog, rain, snow, or at night. To address this, we present Weather-Conditioned Depth Anything (DA-W), a framework that explicitly disentangles style from content for weather-robust depth estimation.

Specifically, we introduce a Style Filter trained on a curated mix of real and synthetic degradation datasets to extract content-independent, degradation-aware weather embeddings. This style embedding is then injected into the Depth Anything backbone using a parameter-efficient, zero-initialized adapter. Such a lightweight modulation allows a single unified model to robustly adapt to diverse conditions—including fog, rain, snow, and low-light—while avoiding catastrophic forgetting of its core generalization abilities in normal conditions. We train the adapter using a pseudo-label distillation and alignment strategy.

Our comprehensive experiments demonstrate that our proposed DA-W achieves state-of-the-art robust depth estimation, improving AbsRel by an average of 3.7% on our curated weather benchmarks, while matching or slightly outperforming performance on standard clean benchmarks.

Depth Visualization on Videos

Snow
Fog
Low-light
Rain

Source footage: Snow and Rain by Kim Dodge, Fog by Talha Uğuz via Pexels; Low-light from Boreas.

🎯 Pipeline 🎯

DA-W pipeline with a weather style filter and zero-initialized adapters in the Depth Anything backbone.
DA-W extracts a degradation-aware weather embedding with a lightweight Style Filter, then injects it into the frozen Depth Anything backbone through parameter-efficient, zero-initialized adapters. A unified model adapts to diverse adverse conditions while preserving its generalization in clean scenes.

♣️ Qualitative Comparison ♣️

Each aligned row shows the input beside a baseline-to-DA-W slider.

♠️ Quantitative Results ♠️

Zero-shot relative depth estimation on real adverse-weather benchmarks.

Zero-shot relative depth results on five real adverse-weather benchmarks; method citation markers are omitted for readability.

Citation

Proceedings metadata will be added when it becomes available.

@inproceedings{xu2026weather,
  title     = {Weather-Conditioned Depth Anything},
  author    = {Xu, Zhaoming and Hu, Chan-Wei and Huang, Kuan-Ru and
               Zhu, Zihao and Li, Renjie and Zhou, Yang and Tu, Zhengzhong},
  booktitle = {European Conference on Computer Vision (ECCV)},
  year      = {2026}
}