在J5芯片复现BEVFormer问题

J5_OE_1.1.74以及J6EM_OE_3.0.22

我们手里目前有地平线J5的板子，前期准备在J5上部署BEVFormer算法时遇到了很多问题。之后在社区看到了将BEVFormer算法部署到J6的帖子，于是计划将J6OE包中的BEVFormer算法的代码复制过来在J5的OE包中使用。

TypeError: HorizonTemporalSelfAttention has not registered in any of registry ['HAT_OBJECT_REGISTRY'] and is not a class, which is not allowed. 

以下是从终端复制的完整报错信息：

Traceback (most recent call last):

 File "/root/.local/lib/python3.8/site-packages/hat/engine/ddp_trainer.py", line 457, in _with_exception

  fn(*args)

 File "/open_explorer/ddk/samples/ai_toolchain/horizon_model_train_sample/scripts/tools/train.py", line 202, in train_entrance

  trainer = build_from_registry(trainer)

 File "/root/.local/lib/python3.8/site-packages/hat/registry.py", line 414, in build_from_registry

  return _impl(x)

 File "/root/.local/lib/python3.8/site-packages/hat/registry.py", line 370, in _impl

  build_x = dict(((key, _impl(value)) for key, value in x.items())) # noqa

 File "/root/.local/lib/python3.8/site-packages/hat/registry.py", line 370, in

  build_x = dict(((key, _impl(value)) for key, value in x.items())) # noqa

 File "/root/.local/lib/python3.8/site-packages/hat/registry.py", line 370, in _impl

  build_x = dict(((key, _impl(value)) for key, value in x.items())) # noqa

 File "/root/.local/lib/python3.8/site-packages/hat/registry.py", line 370, in

  build_x = dict(((key, _impl(value)) for key, value in x.items())) # noqa

 File "/root/.local/lib/python3.8/site-packages/hat/registry.py", line 370, in _impl

  build_x = dict(((key, _impl(value)) for key, value in x.items())) # noqa

 File "/root/.local/lib/python3.8/site-packages/hat/registry.py", line 370, in

  build_x = dict(((key, _impl(value)) for key, value in x.items())) # noqa

 File "/root/.local/lib/python3.8/site-packages/hat/registry.py", line 370, in _impl

  build_x = dict(((key, _impl(value)) for key, value in x.items())) # noqa

 File "/root/.local/lib/python3.8/site-packages/hat/registry.py", line 370, in

  build_x = dict(((key, _impl(value)) for key, value in x.items())) # noqa

 File "/root/.local/lib/python3.8/site-packages/hat/registry.py", line 370, in _impl

  build_x = dict(((key, _impl(value)) for key, value in x.items())) # noqa

 File "/root/.local/lib/python3.8/site-packages/hat/registry.py", line 370, in

  build_x = dict(((key, _impl(value)) for key, value in x.items())) # noqa

 File "/root/.local/lib/python3.8/site-packages/hat/registry.py", line 387, in _impl

  _raise_invalid_type_error(object_type)

 File "/root/.local/lib/python3.8/site-packages/hat/registry.py", line 209, in _raise_invalid_type_error

  raise TypeError(err_msg)

TypeError: HorizonTemporalSelfAttention has not registered in any of registry ['HAT_OBJECT_REGISTRY'] and is not a class, which is not allowed. 

ERROR:__main__:train failed! process 0 terminated with exit code 1

Traceback (most recent call last):

 File "tools/train.py", line 307, in

  raise e

 File "tools/train.py", line 293, in

  train(

 File "tools/train.py", line 274, in train

  launch(

 File "/root/.local/lib/python3.8/site-packages/hat/engine/ddp_trainer.py", line 426, in launch

  mp.spawn(

 File "/usr/local/lib/python3.8/dist-packages/torch/multiprocessing/spawn.py", line 240, in spawn

  return start_processes(fn, args, nprocs, join, daemon, start_method='spawn')

 File "/usr/local/lib/python3.8/dist-packages/torch/multiprocessing/spawn.py", line 198, in start_processes

  while not context.join():

 File "/usr/local/lib/python3.8/dist-packages/torch/multiprocessing/spawn.py", line 149, in join

  raise ProcessExitedException(

torch.multiprocessing.spawn.ProcessExitedException: process 0 terminated with exit code 1