Camera failure test about bevfusion

[kcyu2014]

Thanks for your interest! For the robustness scenarios, we compose a formal robustness benchmark for lidar-camera fusion models. Please follow this repo and paper at https://arxiv.org/abs/2205.14951 for more details. We will release the full benchmark by the end of June 2022.

[alexa19lab]

Thanks for your interest! For the robustness scenarios, we compose a formal robustness benchmark for lidar-camera fusion models. Please follow this repo and paper at https://arxiv.org/abs/2205.14951 for more details. We will release the full benchmark by the end of June 2022.

Thanks for such useful tools and benchmark! May I ask whether the robustness evaluation is based on one model or it is based on different models trained with datasets with different noises? In another word, do we need to keep the consistency between training and evaluation dataset?

[tingtingliangvs]

In this paper, data augmentation is consistent between training and evaluation datasets. You are welcome to train and evaluate models under the combined augmentation strategies and share your findings with us.