【已解决】undefined symbol: _ZN2at4_ops5zeros4callEN3c108ArrayRefIlEENS2_8optionalINS2_10ScalarTypeEEENS5

问题描述

原来的项目在升级了torch和CUDA版本之后，运行原来的代码后报错：

/root/miniconda3/envs/internimage/lib/python3.7/site-packages/mmcv/cnn/bricks/transformer.py:33: UserWarning: Fail to import ``MultiScaleDeformableAttention`` from ``mmcv.ops.multi_scale_deform_attn``, You should install ``mmcv-full`` if you need this module.
  warnings.warn('Fail to import ``MultiScaleDeformableAttention`` from '
Traceback (most recent call last):
  File "image_demo.py", line 7, in <module>
    import mmseg_custom   # noqa: F401,F403
  File "/root/autodl-tmp/InternImage/segmentation/mmseg_custom/__init__.py", line 7, in <module>
    from .models import *  # noqa: F401,F403
  File "/root/autodl-tmp/InternImage/segmentation/mmseg_custom/models/__init__.py", line 7, in <module>
    from .backbones import *  # noqa: F401,F403
  File "/root/autodl-tmp/InternImage/segmentation/mmseg_custom/models/backbones/__init__.py", line 7, in <module>
    from .intern_image import InternImage
  File "/root/autodl-tmp/InternImage/segmentation/mmseg_custom/models/backbones/intern_image.py", line 15, in <module>
    from mmseg.models.builder import BACKBONES
  File "/root/miniconda3/envs/internimage/lib/python3.7/site-packages/mmseg/models/__init__.py", line 2, in <module>
    from .backbones import *  # noqa: F401,F403
  File "/root/miniconda3/envs/internimage/lib/python3.7/site-packages/mmseg/models/backbones/__init__.py", line 7, in <module>
    from .fast_scnn import FastSCNN
  File "/root/miniconda3/envs/internimage/lib/python3.7/site-packages/mmseg/models/backbones/fast_scnn.py", line 7, in <module>
    from mmseg.models.decode_heads.psp_head import PPM
  File "/root/miniconda3/envs/internimage/lib/python3.7/site-packages/mmseg/models/decode_heads/__init__.py", line 2, in <module>
    from .ann_head import ANNHead
  File "/root/miniconda3/envs/internimage/lib/python3.7/site-packages/mmseg/models/decode_heads/ann_head.py", line 8, in <module>
    from .decode_head import BaseDecodeHead
  File "/root/miniconda3/envs/internimage/lib/python3.7/site-packages/mmseg/models/decode_heads/decode_head.py", line 11, in <module>
    from ..losses import accuracy
  File "/root/miniconda3/envs/internimage/lib/python3.7/site-packages/mmseg/models/losses/__init__.py", line 6, in <module>
    from .focal_loss import FocalLoss
  File "/root/miniconda3/envs/internimage/lib/python3.7/site-packages/mmseg/models/losses/focal_loss.py", line 6, in <module>
    from mmcv.ops import sigmoid_focal_loss as _sigmoid_focal_loss
  File "/root/miniconda3/envs/internimage/lib/python3.7/site-packages/mmcv/ops/__init__.py", line 2, in <module>
    from .active_rotated_filter import active_rotated_filter
  File "/root/miniconda3/envs/internimage/lib/python3.7/site-packages/mmcv/ops/active_rotated_filter.py", line 10, in <module>
    ['active_rotated_filter_forward', 'active_rotated_filter_backward'])
  File "/root/miniconda3/envs/internimage/lib/python3.7/site-packages/mmcv/utils/ext_loader.py", line 13, in load_ext
    ext = importlib.import_module('mmcv.' + name)
  File "/root/miniconda3/envs/internimage/lib/python3.7/importlib/__init__.py", line 127, in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
ImportError: /root/miniconda3/envs/internimage/lib/python3.7/site-packages/mmcv/_ext.cpython-37m-x86_64-linux-gnu.so: undefined symbol: _ZN2at4_ops5zeros4callEN3c108ArrayRefIlEENS2_8optionalINS2_10ScalarTypeEEENS5_INS2_6LayoutEEENS5_INS2_6DeviceEEENS5_IbEE

可以看到最后一行错误：

ImportError: /root/miniconda3/envs/internimage/lib/python3.7/site-packages/mmcv/_ext.cpython-37m-x86_64-linux-gnu.so: undefined symbol: _ZN2at4_ops5zeros4callEN3c108ArrayRefIlEENS2_8optionalINS2_10ScalarTypeEEENS5_INS2_6LayoutEEENS5_INS2_6DeviceEEENS5_IbEE

解决方法

我的mmcv版本较老，是1.5.0，去官方文档查看了后发现要安装对应torch和cuda版本的mmcv-full

替换成我对应的版本：

 pip install mmcv-full==1.5.0 -f https://download.openmmlab.com/mmcv/dist/cu117/torch1.13.0/index.html               

注意要先卸载之前的mmcv-full，再安装这个，不然会提示已安装

问题解决！！！