Skip to content

Latest commit

 

History

History

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 

Sparse4D

To enable training in AMD ROCm platform, small modifications are needed:

  1. Prepare data
  2. Compile the deformable_aggregation CUDA op:
cd projects/mmdet3d_plugin/ops
python3 setup.py develop
  1. update motmetrics
pip install motmetrics==1.1.3

将 /usr/local/lib/python3.10/dist-packages/motmetrics/metrics.py 中
from collections import OrderedDict, Iterable
修改为
from collections import OrderedDict
try:
    from collections.abc import Iterable
except ImportError:
    from collections import Iterable

  1. pip install pandas==1.1.5
  2. Download pre-trained weights and Training
bash local_train.sh sparse4dv3_temporal_r50_1x8_bs6_256x704

We have also provided an example to provide out-of-the-box experience.