[ICRA2024] Towards Motion Forecasting with Real-World Perception Inputs: Are End-to-End Approaches Competitive? This is the official implementation of the evaluation protocol proposed in this work for motion forecasting models with real-world perception inputs.
I saw that when you calculated the (MissRate) MR, you set miss_rate_threshold=4.0. Could you tell me why you chose 4.0 as a threshold? Or are there any References?
The reason why the numbers for ViP3D and UniAD in Table 1 differ from those in the papers of ViP3D and UniAD.
In the case of UniAD, whether the learning and finetuning included training the occformer and planner as well.
When measuring the motion forecasting metric, how was the average calculated for categories, and whether it included averaging together vehicles (car, truck, construction vehicle, bus, trailer, motorcycle, and bicycle) or pedestrians as well.