In this Argo AI sponsored research project, we explore multimodal object detection for autonomous driving. Specifically, we focus on detection with RGB and infrared signals in the driving scenes.
It is well known that infrared images are more robust to lighting (e.g., dynamic range and night), weather conditions (e.g., fog and snow), etc. Therefore, the joint force from RGB and infrared images allows for potential opportunities for better object detection. Arising in this study, there are novel challenges, solving which will be of great value to the community, such as how to train a better model with unaligned RGB and infrared images. We also expect to collect a large-scale dataset to release, along with our codebase to foster broad research in the community.