Publications

SWIR imaging represents a promising approach for enhancing road perception in adverse conditions, though it remains constrained by the scarcity of annotated data compared to models pre-trained in the visible spectrum. This paper presents a comparative study of adaptation strategies for YOLOv8 in the context of multi-class road infrared detection. We demonstrate that partial network freezing offers an efficiency-performance trade-off comparable to full fine-tuning, while highlighting that a structural performance gap consistently persists in favor of the visible domain. Finally, evaluating image synthesis through image-to-image translation and training on mixed datasets reveals effective strategies for mitigating the lack of real-world data, although these methods do not fully bridge the gap between modalities.

@InProceedings{Mehra_2025_ICCV,
    author    = {Mehra, Rohan and Riffard, Alexandre and Labussière, Mathieu and Duthon, Pierre and Aufrère, Romuald},
    title     = {Would SWIR modality help for detection and segmentation in harsh weather conditions? An experimental study.},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops},
    month     = {October},
    year      = {2025},
    pages     = {2211-2219}
}