Windows11下DeepLabCut 3.0.0rc4环境启动失败求助

环境与安装信息 #

• 系统：Windows11
• Python版本：3.10.14
• TensorFlow版本：2.10
• DeepLabCut版本：3.0.0rc4
• 运行方式：本地GPU

安装命令：

conda env create -f DEEPLABCUT.yaml
conda activate deeplabcut
python -m deeplabcut 

DEEPLABCUT.yaml配置：

name: DEEPLABCUT
channels:
  - conda-forge
  - defaults
dependencies:
  - python=3.10
  - pip
  - ipython
  - jupyter
  - nb_conda
  - notebook<7.0.0
  - ffmpeg
  - pytables==3.8.0
  - git
  - pip:
    - "git+https://github.com/DeepLabCut/DeepLabCut.git@pytorch_dlc#egg=deeplabcut[gui,modelzoo,wandb]"

初始启动错误 #

Traceback (most recent call last):
  File "C:\ProgramData\anaconda3\envs\deeplabcut\lib\runpy.py", line 187, in _run_module_as_main
    mod_name, mod_spec, code = _get_module_details(mod_name, _Error)
  File "C:\ProgramData\anaconda3\envs\deeplabcut\lib\runpy.py", line 146, in _get_module_details
    return _get_module_details(pkg_main_name, error)
  File "C:\ProgramData\anaconda3\envs\deeplabcut\lib\runpy.py", line 110, in _get_module_details
    __import__(pkg_name)
  File "C:\ProgramData\anaconda3\envs\deeplabcut\lib\site-packages\deeplabcut\__init__.py", line 76, in <module>
    from deeplabcut.pose_tracking_pytorch import transformer_reID
  File "C:\ProgramData\anaconda3\envs\deeplabcut\lib\site-packages\deeplabcut\pose_tracking_pytorch\__init__.py", line 14, in <module>
    from .train_dlctransreid import train_tracking_transformer
  File "C:\ProgramData\anaconda3\envs\deeplabcut\lib\site-packages\deeplabcut\pose_tracking_pytorch\train_dlctransreid.py", line 15, in <module>
    import torch
  File "C:\ProgramData\anaconda3\envs\deeplabcut\lib\site-packages\torch\__init__.py", line 148, in <module>
    raise err
OSError: [WinError 126] The specified module could not be found. Error loading "C:\ProgramData\anaconda3\envs\deeplabcut\lib\site-packages\torch\lib\fbgemm.dll" or one of its dependencies.

尝试修复后的问题 #

卸载torch-2.4.0+cu118并安装torch==2.3.1后，出现依赖冲突：

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
torchaudio 2.4.0+cu118 requires torch==2.4.0+cu118, but you have torch 2.3.1 which is incompatible.
torchvision 0.19.0 requires torch==2.4.0, but you have torch 2.3.1 which is incompatible.

同时触发新的启动错误：

Loading DLC 3.0.0rc4...
Traceback (most recent call last):
  File "C:\ProgramData\anaconda3\envs\deeplabcut\lib\runpy.py", line 187, in _run_module_as_main
    mod_name, mod_spec, code = _get_module_details(mod_name, _Error)
  File "C:\ProgramData\anaconda3\envs\deeplabcut\lib\runpy.py", line 146, in _get_module_details
    return _get_module_details(pkg_main_name, error)
  File "C:\ProgramData\anaconda3\envs\deeplabcut\lib\runpy.py", line 110, in _get_module_details
    __import__(pkg_name)
  File "C:\ProgramData\anaconda3\envs\deeplabcut\lib\site-packages\deeplabcut\__init__.py", line 61, in <module>
    from deeplabcut.modelzoo.video_inference import video_inference_superanimal
  File "C:\ProgramData\anaconda3\envs\deeplabcut\lib\site-packages\deeplabcut\modelzoo\video_inference.py", line 23, in <module>
    from deeplabcut.pose_estimation_pytorch.modelzoo.train_from_coco import adaptation_train
  File "C:\ProgramData\anaconda3\envs\deeplabcut\lib\site-packages\deeplabcut\pose_estimation_pytorch\__init__.py", line 11, in <module>
    from deeplabcut.pose_estimation_pytorch.apis import (
  File "C:\ProgramData\anaconda3\envs\deeplabcut\lib\site-packages\deeplabcut\pose_estimation_pytorch\apis\__init__.py", line 12, in <module>
    from deeplabcut.pose_estimation_pytorch.apis.analyze_images import (
  File "C:\ProgramData\anaconda3\envs\deeplabcut\lib\site-packages\deeplabcut\pose_estimation_pytorch\apis\analyze_images.py", line 27, in <module>
    from deeplabcut.pose_estimation_pytorch.apis.utils import (
  File "C:\ProgramData\anaconda3\envs\deeplabcut\lib\site-packages\deeplabcut\pose_estimation_pytorch\apis\utils.py", line 23, in <module>
    from deeplabcut.pose_estimation_pytorch.data.dataset import PoseDatasetParameters
  File "C:\ProgramData\anaconda3\envs\deeplabcut\lib\site-packages\deeplabcut\pose_estimation_pytorch\data\__init__.py", line 11, in <module>
    from deeplabcut.pose_estimation_pytorch.data.base import Loader
  File "C:\ProgramData\anaconda3\envs\deeplabcut\lib\site-packages\deeplabcut\pose_estimation_pytorch\data\base.py", line 20, in <module>
    from deeplabcut.pose_estimation_pytorch.data.dataset import (
  File "C:\ProgramData\anaconda3\envs\deeplabcut\lib\site-packages\deeplabcut\pose_estimation_pytorch\data\dataset.py", line 20, in <module>
    from deeplabcut.pose_estimation_pytorch.data.utils import (
  File "C:\ProgramData\anaconda3\envs\deeplabcut\lib\site-packages\deeplabcut\pose_estimation_pytorch\data\utils.py", line 21, in <module>
    from torchvision.ops import box_convert
  File "C:\ProgramData\anaconda3\envs\deeplabcut\lib\site-packages\torchvision\__init__.py", line 10, in <module>
    from torchvision import _meta_registrations, datasets, io, models, ops, transforms, utils  # usort:skip
  File "C:\ProgramData\anaconda3\envs\deeplabcut\lib\site-packages\torchvision\_meta_registrations.py", line 163, in <module>
    @torch.library.register_fake("torchvision::nms")
AttributeError: module 'torch.library' has no attribute 'register_fake'

解决方案 #

1. 统一Torch套件版本 #

错误核心是torch、torchvision、torchaudio版本不兼容：torch.library.register_fake是Torch 2.4+新增API，而Torch 2.3.1无此方法；同时torchvision 0.19.0强制要求Torch 2.4.0。需安装完全匹配的GPU版本套件：

激活环境后执行：

pip uninstall -y torch torchvision torchaudio
pip install torch==2.3.1 torchvision==0.18.1 torchaudio==2.3.1 --index-url https://download.pytorch.org/whl/cu118

2. 修复fbgemm.dll依赖 #

若安装后仍出现fbgemm.dll加载错误：

• 安装最新版Microsoft Visual C++ Redistributable（x64）
• 确保NVIDIA驱动适配CUDA 11.8（建议更新到450.80.02及以上版本）

3. 验证环境与启动 #

执行以下代码验证GPU可用性：

import torch
print(torch.cuda.is_available())

输出True则GPU配置正常，重新启动DeepLabCut：

python -m deeplabcut

4. 备选：使用官方推荐的conda配置 #

若上述方法无效，删除现有环境，使用以下yaml创建新环境：

name: dlc_gpu
channels:
  - conda-forge
  - defaults
dependencies:
  - python=3.10
  - pip
  - ipython
  - jupyter
  - nb_conda
  - notebook<7.0.0
  - ffmpeg
  - pytables==3.8.0
  - git
  - cudatoolkit=11.8
  - pip:
    - deeplabcut[gui,modelzoo,wandb]==3.0.0rc4
    - torch==2.3.1 torchvision==0.18.1 torchaudio==2.3.1 --index-url https://download.pytorch.org/whl/cu118

创建并启动环境：

conda env create -f dlc_gpu.yaml
conda activate dlc_gpu
python -m deeplabcut

内容的提问来源于stack exchange，提问作者Hannah Lyons