一、在nvidia和cuda安装成功的情况下,运行train.py时出现以下报错:
- RuntimeError: Missing libdevice file
- ImportError: libcupti.so.11.7: cannot open shared object file: No such file or directory
- [Errno 2] No such file or directory: ‘:/usr/local/cuda-11.8/bin/nvcc’
- Unrecognized tensor type ID: AutocastCUDA
- ERROR: Command errored out with exit status 1:
- libNVVM cannot be found. Do
conda install cudatoolkit
: libnvvm.so: cannot - ValueError: Unknown CUDA arch or GPU not supported
这些问题都是由于cuda环境配置没有配置完全,下面是配置方法:
1、打开~/.bashrc文件
vim ~/.bashrc
2、在文件末尾输入:
export PATH=/usr/local/cuda-11.8/bin:$PATH
export CUDA_HOME=/usr/local/cuda-11.8
export LD_LIBRARY_PATH=/usr/local/cuda-11.8/lib64
export NUMBAPRO_NVVM=/usr/local/cuda-11.8/nvvm/lib64/libnvvm.so
export NUMBAPRO_LIBDEVICE=/usr/local/cuda-11.8/nvvm/libdevice/
3、让配置生效
source ~/.bashrc
二、nuScenes数据集太大,只使用其中一个v1.0-trainval01_blobs.tgz(part 1)进行训练,运行train.py时出现以下报错:\
FileNotFoundError: [Errno 2] No such file or directory: ‘…/data/nuscenes/v1.0-trainval/samples/LIDAR_TOP/n008-2018-08-01-15-16-36-0400__LIDAR_TOP__1533151603547590.pcd.bin’
原因:生成的val.pkl是针对850个场景,所有样本的,在评估的时候会报错:在samples找不到某个数据
解决办法:进入nuscenes.utils.py,在374行进行以下修改:
# if sample['scene_token'] in train_scenes:
# train_nusc_infos.append(info)
# else:
# val_nusc_infos.append(info)
if sample['scene_token'] in train_scenes:
train_nusc_infos.append(info)
elif sample['scene_token'] in val_scenes:
val_nusc_infos.append(info)
else:
pass