Bug when running the testing script

[db.chen]

File "FB-BEV/mmdet3d/models/fbbev/detectors/fbocc.py", line 486, in forward_test
if num_augs==1 and not img_metas[0][0].get('tta_config', dict(dist_tta=False))['dist_tta']:
TypeError: 'DataContainer' object is not subscriptable

[Herman]

I also encountered this error, have you solved this problem?

[Egozjuer]

I encounted the same question, @zhiqi-li please teel us the reasons

[Zhiqi Li]

I think this is the problem of versions of openmmlab packages. The input img_metas still are wrapped with its DataContainer, and this can be solved by calling .data property of DataContainer to recover the raw data.

[Egozjuer]

mmcls 0.25.0
mmcv-full 1.5.2
mmdet 2.24.0
mmdet3d 1.0.0rc4
mmsegmentation 0.24.0
this is my envs,actually it's created according to your install.md,I have tried the following operation:
type(img_metas[0])
<class 'mmcv.parallel.data_container.DataContainer'>
img_metas[0].data()
Traceback (most recent call last):
File "", line 1, in
TypeError: 'list' object is not callable

As you can see,.data() seems not to be useful for the question @zhiqi-li

[db.chen]

mmcls 0.25.0 mmcv-full 1.5.2 mmdet 2.24.0 mmdet3d 1.0.0rc4 mmsegmentation 0.24.0 this is my envs,actually it's created according to your install.md,I have tried the following operation: type(img_metas[0]) <class 'mmcv.parallel.data_container.DataContainer'> img_metas[0].data() Traceback (most recent call last): File "", line 1, in TypeError: 'list' object is not callable

As you can see,.data() seems not to be useful for the question @zhiqi-li

I change the code as this and it works:

# if num_augs==1 and not img_metas[0][0].get('tta_config', dict(dist_tta=False))['dist_tta']:
if num_augs==1 and not img_metas[0].data[0][0].get('tta_config', dict(dist_tta=False))['dist_tta']:
    # return self.simple_test(points[0], img_metas[0], img_inputs[0],
    #                     **kwargs)
    return self.simple_test(points[0], img_metas[0].data[0], img_inputs[0],
                        **kwargs)
else:

[Xinhao Liu]

@cdb342 Maybe you should consider submitting a pull request that addresses this problem.