[Fix] Fix the device mismatch error when torch version<1.11

[Ginray]

Motivation

When the torch version<1.11, an error will be reported when executing the centerpoint model:
The error message is:
RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu!

This happens at centerpoint_head.py line542, due to center_int is saved on cuda, but feature_map_size is saved on the CPU.

if not (0 <= center_int[0] < feature_map_size[0]
        and 0 <= center_int[1] < feature_map_size[1]):
    continue

A more intuitive example:

>>> import torch
>>> torch.__version__
'1.8.0'
>>> a = torch.tensor([1])
>>> b = torch.tensor([2]).to('cuda')
>>> a<b
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu!

Modification

Make center_int and feature_map_size saved on the same device to avoid errors in the lower version of the torch.

BC-breaking (Optional)

Not involved

Use cases (Optional)

Not involved

[codecov]

Codecov Report

Base: 46.37% // Head: 46.96% // Increases project coverage by +0.59% 🎉

Coverage data is based on head (7f11980) compared to base (d84e081).
Patch coverage: 66.37% of modified lines in pull request are covered.

❗ Current head 7f11980 differs from pull request most recent head 52f9205. Consider uploading reports for the commit 52f9205 to get more accurate results

Additional details and impacted files

@@             Coverage Diff             @@
##           dev-1.x    #2308      +/-   ##
===========================================
+ Coverage    46.37%   46.96%   +0.59%     
===========================================
  Files          259      262       +3     
  Lines        21390    21964     +574     
  Branches      3379     3460      +81     
===========================================
+ Hits          9919    10315     +396     
- Misses       10827    10987     +160     
- Partials       644      662      +18     

Flag	Coverage Δ
unittests	46.96% <66.37%> (+0.59%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
mmdet3d/datasets/transforms/__init__.py	100.00% <ø> (ø)
mmdet3d/evaluation/__init__.py	100.00% <ø> (ø)
mmdet3d/models/dense_heads/centerpoint_head.py	21.77% <0.00%> (ø)
mmdet3d/models/dense_heads/fcaf3d_head.py	16.23% <ø> (ø)
mmdet3d/models/detectors/mink_single_stage.py	43.58% <ø> (ø)
mmdet3d/testing/data_utils.py	70.83% <ø> (ø)
mmdet3d/models/voxel_encoders/voxel_encoder.py	38.46% <6.94%> (-12.08%)	⬇️
...t3d/models/data_preprocessors/data_preprocessor.py	39.10% <17.24%> (-5.05%)	⬇️
mmdet3d/datasets/transforms/formating.py	58.03% <33.33%> (+1.78%)	⬆️
mmdet3d/models/data_preprocessors/voxelize.py	33.92% <33.92%> (ø)
... and 28 more

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.