MonoWAD: A Smart Way to Detect 3D Objects in Any Weather

Detecting 3D objects from a single camera is tough, especially when the weather isn't perfect. Most current methods work great on clear days but struggle with fog. That's where MonoWAD comes in. It's a clever system that can handle different weather conditions. MonoWAD has two parts: a weather codebook and a weather-adaptive diffusion model. The codebook stores info about clear weather and creates a weather-reference feature for any input. The diffusion model then uses this reference to enhance the input feature. It's like having a smart helper that tells you how much improvement is needed based on the weather.

To make sure the feature representation is good, MonoWAD uses a weather-adaptive enhancement loss. This loss helps improve features under both clear and foggy conditions. Tests showed that MonoWAD can detect 3D objects robustly in various weather conditions. You can find the code and dataset online.

Actions