Closed wutheringcoo closed 6 months ago
We have not tried to deploy sparse4d to orin, only to the horizon journey series chips. There is not much difference in inference speed between sparse4d and sparse3d (single-frame model), and sparse3d cannot achieve end-to-end tracking. See the paper for the inference speed of the GPU and the performance of the single-frame model.
Hi, Do you test latency of the model with config: Res50 + 256x704? if the speed is too slowly, how to change sparse4D to sparse3d(it can alse support end2end track&detection), and performance metric will drop significantly?