Google Colab中FastAI图像增强转换时出现RuntimeError求助

This error is almost certainly due to version incompatibilities between FastAI and PyTorch in Google Colab's default environment. Colab often auto-updates its pre-installed libraries, which can break code that worked with older versions (your local notebook works because it's using the stable, compatible versions you had set up).

Fix Steps: #

1. Pin FastAI and PyTorch to compatible versions
   Add this code at the very top of your Colab notebook, run it first, then restart the runtime when prompted:

   !pip install fastai==1.0.61 torch==1.4.0 torchvision==0.5.0
   

   These versions are known to work well together and should resolve the tensor type mismatch in affine transformations.

2. Verify version compatibility
   After restarting, run this to confirm you have the correct versions:

   import fastai
   import torch
   print(f"FastAI version: {fastai.__version__}")
   print(f"PyTorch version: {torch.__version__}")
   

   Ensure the output matches 1.0.61 for FastAI and 1.4.0 for PyTorch.

3. Re-run your original code
   Once the versions are locked, your data block code should execute without the RuntimeError related to Float/Double types.

Why This Happened: #

The error occurs during the affine transformation step in your data augmentation (max_rotate=360 triggers this). Newer PyTorch versions handle tensor types differently, and older FastAI versions (like the one used in that Kaggle notebook) aren't compatible with these changes. By pinning to the versions that were stable when the notebook was created, you eliminate this type mismatch issue.

内容的提问来源于stack exchange，提问作者Shreya Pandita