Supplementary Material N-ImageNet: Towards Robust, Fine-Grained Object Recognition with Event Cameras

We report the dimensions of the hardware used for recording N-ImageNet data in Figure A.1. The custom hardware is utilized for generating N-ImageNet along with its variants, where the detailed generation process is described in Section 3.1. In addition, we provide a sample Arduino [2] script for generating arbitrary event camera motion in Figure A.2. Using the aforementioned setup, we make 50μs event camera recordings of each ImageNet [14] image.