No module named 'mmcv._ext' 怎么解决

[gangxu822]

from mmpose.models.builder import build_pose_estimator

File "/opt/nas/p/conda/envs/MuseTalk/lib/python3.10/site-packages/mmpose/models/init.py", line 8, in
from .heads import * # noqa
File "/opt/nas/p/conda/envs/MuseTalk/lib/python3.10/site-packages/mmpose/models/heads/init.py", line 11, in
from .transformer_heads import EDPoseHead
File "/opt/nas/p/conda/envs/MuseTalk/lib/python3.10/site-packages/mmpose/models/heads/transformer_heads/init.py", line 2, in
from .edpose_head import EDPoseHead
File "/opt/nas/p/conda/envs/MuseTalk/lib/python3.10/site-packages/mmpose/models/heads/transformer_heads/edpose_head.py", line 14, in
from mmcv.ops import MultiScaleDeformableAttention
File "/opt/nas/p/conda/envs/MuseTalk/lib/python3.10/site-packages/mmcv/ops/init.py", line 3, in
from .active_rotated_filter import active_rotated_filter
File "/opt/nas/p/conda/envs/MuseTalk/lib/python3.10/site-packages/mmcv/ops/active_rotated_filter.py", line 10, in
ext_module = ext_loader.load_ext(
File "/opt/nas/p/conda/envs/MuseTalk/lib/python3.10/site-packages/mmcv/utils/ext_loader.py", line 13, in load_ext
ext = importlib.import_module('mmcv.' + name)
File "/opt/nas/p/conda/envs/MuseTalk/lib/python3.10/importlib/init.py", line 126, in import_module
return _bootstrap._gcd_import(name[level:], package, level)
ModuleNotFoundError: No module named 'mmcv._ext'

[AhmadAFS1]

If you are running on mac, there is a repo which showcases how to run this.
When I tried it was instanely slow though on macos, since this repo doesn't support MPS too well. Most of this repo requires CUDA.

https://github.com/rrfunde/MuseTalk/blob/61ef17f0fc5b522ef660ac08a9b81de374b630e5/MACOS_COMPATIBILITY_NOTE.md?plain=1#L205

[hanee-shousha]

I faced the same error you have above on a GPU machine with CUDA and I was able to solve it by downgrading the mmcv package and used this version instead of 2.1.0:

mim install "mmcv==2.0.1"

[ajeema]

I solved it by building from source

[chenyujie2580]

mim install "mmcv==2.0.1"

not work

[chenyujie2580]

I solved it by building from source

not work

[ajeema]

try this .sh script
#!/usr/bin/env bash
set -e

ENV_NAME=nerfstream

echo "============================================"
echo " Fixing MuseTalk inside conda env: $ENV_NAME"
echo " Updated working configuration - all versions tested"
echo " PyTorch 2.0.1+cu117, mmcv 2.1.0, mmpose 1.2.0"
echo "============================================"

---------------------------------------------

1. Activate existing environment

---------------------------------------------

#source ~/miniforge/etc/profile.d/conda.sh
#conda activate $ENV_NAME

---------------------------------------------

2. Pin pip and upgrade build tools

---------------------------------------------

python -m pip install --upgrade pip setuptools wheel

---------------------------------------------

3. Install PyTorch 2.0.1 with CUDA 11.7

This version is compatible with mmcv pre-built wheels

---------------------------------------------

pip uninstall -y torch torchvision torchaudio || true
pip install torch==2.0.1 torchvision==0.15.2 torchaudio==2.0.2
--index-url https://download.pytorch.org/whl/cu117

python - << 'EOF'
import torch
assert torch.cuda.is_available(), "CUDA NOT AVAILABLE"
print("Torch:", torch.version)
print("CUDA:", torch.version.cuda)
print("GPUs:", torch.cuda.device_count())
EOF

---------------------------------------------

4. Install OpenMMLab packages via mim

---------------------------------------------

pip install --no-cache-dir -U openmim

Remove any conflicting packages first

pip uninstall -y mmdet mmcv mmcv-full mmpose mmengine chumpy || true

Install mmengine (must be 0.9.1 for mmcv 2.1.0 compatibility)

mim install mmengine==0.9.1

Install mmcv 2.1.0 with pre-built wheel for CUDA 11.7 + PyTorch 2.0.1

Note: Using PyTorch 2.0.0 wheels which are compatible with 2.0.1

mim install "mmcv>=2.0.0,<2.2.0"
-f https://download.openmmlab.com/mmcv/dist/cu117/torch2.0.0/index.html

Install mmdet (3.3.0 is compatible with mmcv 2.1.0)

mim install "mmdet>=3.1.0,<3.4.0"

Install mmpose 1.2.0 (tested and working version)

pip install mmpose==1.2.0

---------------------------------------------

6. Verify OpenMMLab stack

---------------------------------------------

python - << 'EOF'
import mmengine, mmcv, mmpose
from mmpose.apis import inference_topdown, init_model
print("MMEngine:", mmengine.version)
print("MMCV:", mmcv.version)
print("MMPose:", mmpose.version)
print("Checking mmcv._ext...")
import mmcv._ext
print("mmcv._ext OK!")
EOF

---------------------------------------------

7. Install MuseTalk dependencies with compatible versions

---------------------------------------------

Install compatible versions of huggingface ecosystem

pip install
"diffusers>=0.20.0,<0.21.0"
"huggingface-hub>=0.16.0,<0.18.0"
"transformers>=4.30.0,<4.35.0"
"accelerate>=0.20.0,<0.25.0"

Fix filelock if corrupted (common issue)

pip uninstall -y filelock || true
pip install "filelock>=3.20.0"

Install other MuseTalk dependencies

pip install
sentencepiece
mediapipe
facexlib
basicsr
einops
librosa
scipy
opencv-python
tqdm
rich

Fix numpy version (required for xtcocotools compatibility)

pip install --force-reinstall "numpy==1.26.4" || true

Optional but recommended

pip uninstall -y tensorflow || true

---------------------------------------------

8. Final sanity check

---------------------------------------------

python - << 'EOF'
import torch
import mmcv
import mmpose

Check PyTorch CUDA

print("PyTorch:", torch.version)
print("CUDA available:", torch.cuda.is_available())
print("CUDA version:", torch.version.cuda)
assert torch.cuda.is_available(), "CUDA NOT AVAILABLE"

Check mmcv._ext

print("MMCV:", mmcv.version)
import mmcv._ext
print("mmcv._ext: OK")

Check mmpose APIs

print("MMPose:", mmpose.version)
from mmpose.apis import inference_topdown, init_model
print("mmpose APIs: OK")

Check diffusers

from diffusers import AutoencoderKL
print("diffusers: OK")

print("\n✅ All critical dependencies verified!")
EOF

echo "============================================"
echo " Installation complete!"
echo " Environment '$ENV_NAME' is MuseTalk-ready"
echo " Run: conda activate $ENV_NAME"
echo " Then: python genavatar_musetalk.py --help"
echo "============================================"