AUTOSIGNAL
Human-curated. Expert-annotated. Every signal traced to source.
人工精选 · 专家点评 · 每条信号可溯源
Research & IP研究与专利
3DarkFusion: 3D-Aware Neural Rendering Fuses Noisy RGB and NIR for Robust Low-Light Imaging3DarkFusion：3D感知神经渲染融合噪声RGB与近红外，实现鲁棒暗光成像
S 1.7
T1
1 sources1 个来源
—
- Researchers propose 3DarkFusion, which introduces 3D-aware neural modeling to fuse extremely noisy RGB with Near-Infrared (NIR) imagery via neural rendering in 3D space, requiring no clean RGB supervision during training.
- Unlike prior RGB-NIR low-light methods that depend on carefully curated paired training data and show limited robustness across different scenarios, 3DarkFusion implicitly fuses the two modalities and stays robust even under severe noise interference.
- Robust low-light imaging from fused RGB-NIR sensing bears on nighttime perception for automotive camera systems, where sensor noise in the dark is a persistent challenge for ADAS and autonomous-driving vision pipelines.
- 研究人员提出3DarkFusion，引入3D感知神经建模，通过神经渲染在3D空间中融合极度噪声的RGB与近红外(NIR)图像，训练过程无需干净的RGB监督数据。
- 与依赖精心构建配对训练数据、且在不同场景下鲁棒性有限的既有RGB-NIR暗光方法不同，3DarkFusion可隐式融合两种模态，在严重噪声干扰下仍保持鲁棒性。
- RGB-NIR融合带来的鲁棒暗光成像，与车载摄像头系统的夜间感知密切相关——暗光下的传感器噪声一直是ADAS及自动驾驶视觉系统面临的持续挑战。