Problems about training

[ChengkaiYang]

Hi,Doctor Zhou:
Well done work!But I have some problems.
1.What causes the OOM problem during training?Recently,someone says that Argoverse2 dataset has different size of map representation input.When training,random combination of large size of samples in a batch may leads to OOM problem.So how can I deal with the problem with a limited memory of GPU.
2.In one of QCNet improvement,after transpose and rotation invariance processing,does each agent in each timestep has a local reference and each agent in each timestep can be regarded as a node in GAT?The edge is the relative position relations of two agents' local reference?I noticed that the relative position relations of two agents' local reference is represented by the relative location,relative direction,relative orientation and relative time-order.I wonder if the four items can uniquely determined the relative relations of two local references?

[ZikangZhou]

1. Perhaps you can modify the data preprocessing. For example, if the number of map nodes is too large, you can drop some map nodes that are too far away from any agents to be predicted.

2. Yes, you can understand the model in this way. The "edge" should preserve the relative information between a pair of nodes such that given one node's absolute position, we are able to recover another node's absolute position via the edge information.

[ChengkaiYang]

Thank you,Doctor Zhou.I also want to ask what is FHE and AHE?

[SunHaoOne]

Thank you,Doctor Zhou.I also want to ask what is FHE and AHE?

I guess they are final heading error and average heading error.