How to fix "ERROR: Failed building wheel for pytorch3d" error on Colab?

Question

I'm trying to use a very cool machine learning Colab. https://colab.research.google.com/drive/1eQLZrNYRZMo9zdnGGccE0hFswGiinO-Z?usp=sharing Running their steps as is, I keep getting ERROR: Failed building wheel for pytorch3d.

After much Googling, I've tried for instance replacing the install line with

!pip install 'git+https://github.com/facebookresearch/pytorch3d.git@stable'

and also

!pip install pytorch3d

The former doesn't work. The latter makes another issue arise:

"ImportError: /usr/local/lib/python3.6/dist-packages/pytorch3d/_C.cpython-36m-x86_64-linux-gnu.so: undefined symbol: _ZNK2at6Tensor7is_cudaEv"

I also tried doing !pip install torch==1.6.0 which causes

Traceback (most recent call last):
  File "./meshrcnn/demo/demo.py", line 11, in <module>
    from detectron2.data import MetadataCatalog
  File "/usr/local/lib/python3.6/dist-packages/detectron2/data/__init__.py", line 4, in <module>
    from .build import (
  File "/usr/local/lib/python3.6/dist-packages/detectron2/data/build.py", line 14, in <module>
    from detectron2.structures import BoxMode
  File "/usr/local/lib/python3.6/dist-packages/detectron2/structures/__init__.py", line 6, in <module>
    from .keypoints import Keypoints, heatmaps_to_keypoints
  File "/usr/local/lib/python3.6/dist-packages/detectron2/structures/keypoints.py", line 6, in <module>
    from detectron2.layers import interpolate
  File "/usr/local/lib/python3.6/dist-packages/detectron2/layers/__init__.py", line 3, in <module>
    from .deform_conv import DeformConv, ModulatedDeformConv
  File "/usr/local/lib/python3.6/dist-packages/detectron2/layers/deform_conv.py", line 10, in <module>
    from detectron2 import _C
ImportError: /usr/local/lib/python3.6/dist-packages/detectron2/_C.cpython-36m-x86_64-linux-gnu.so: undefined symbol: _ZN6caffe28TypeMeta21_typeMetaDataInstanceISt7complexIdEEEPKNS_6detail12TypeMetaDataEv

Have done !pip install mmcv-full===1.2.1 -f https://download.openmmlab.com/mmcv/dist/cu101/torch1.7.0/index.html to try fixing that and the error persists.

Does anyone have ideas for how to make the Colab environment work?

Answer 1

I got the same error and found your post. I tried to install current master (with:

!pip install 'git+https://github.com/facebookresearch/pytorch3d.git'

and after downloading and compiling pytorch3D==0.4.0 it actually worked correctly. I think that what solved the problem was downloading and compiling pytorch3d, so that it got properly linked with cuda: if it downloads a precompiled wheel you'll probably get that undefined symbol error.

Only other change I made: I had to remove a demo=True from the call to group_keypoints() in the bounding rectangle calculation cell (it said that parameter was unknown)

Answer 2

I found the same problem for Google Colab. Then I solved it from https://github.com/facebookresearch/pytorch3d/blob/main/INSTALL.md

In this area, you can see the required solution for your current machine. If you face a problem with Google Colab you can try this code. In general, from inside IPython, or in Google Colab or a Jupyter notebook, you can install with

import sys
import torch
pyt_version_str=torch.__version__.split("+")[0].replace(".", "")
version_str="".join([
    f"py3{sys.version_info.minor}_cu",
    torch.version.cuda.replace(".",""),
    f"_pyt{pyt_version_str}"
])
!pip install --no-index --no-cache-dir pytorch3d -f https://dl.fbaipublicfiles.com/pytorch3d/packaging/wheels/{version_str}/download.html

You have to restart your runtime environment after installing this. In my case, it worked perfectly in Google Colab with GPU instance.