fastai v2.3.1中show_batch与show_results函数无输出问题求助

I built a model using FastAI v2.3.1, but the show_batch() and show_results() functions aren't displaying any content. The model itself works normally during training, but these visualization functions fail to render output. Here's my code:

from fastai.vision.all import *
from fastai.data.all import *
import fastai.vision
import zipfile as zf
import random
import timeit

fields = DataBlock(blocks=(ImageBlock, CategoryBlock),
                  get_items=get_image_files,
                  get_y=yer,
                  splitter=RandomSplitter(valid_pct=0.2, seed=random.randint(0, 10)),
                  item_tfms=RandomResizedCrop(224, min_scale=0.5),
                  batch_tfms=aug_transforms() )
dls = fields.dataloaders(os.path.join(Path(os.getcwd()), "train"), num_workers=0, bs=32)
dls.show_batch()

learn = cnn_learner(dls, resnet18, metrics=error_rate)
learn.fine_tune(2)
learn.show_results()

I've run into similar visualization hiccups with FastAI v2 before, so let's walk through practical fixes and checks to get those previews working:

1. Force Matplotlib to Render Output #

FastAI relies on Matplotlib for visualization, and sometimes Jupyter environments don't auto-trigger the render. Try these steps:

• Add this line at the very top of your notebook to enable inline rendering:

  %matplotlib inline
  

  If you're using JupyterLab, swap it for %matplotlib widget for interactive plots.
• Explicitly call plt.show() right after each visualization function to force the plot to display:

  import matplotlib.pyplot as plt
  
  dls.show_batch()
  plt.show()
  
  learn.show_results()
  plt.show()
  

2. Verify Your Label Extraction Function (yer) #

Your code uses get_y=yer but doesn't include its definition. While you mentioned the model trains fine, a flaky label function could still break visualization. Double-check:

• Does yer return valid string labels? If your data is organized in class-named folders, use FastAI's built-in parent_label to rule out custom function issues:

  fields = DataBlock(blocks=(ImageBlock, CategoryBlock),
                    get_items=get_image_files,
                    get_y=parent_label,  # Replace yer with built-in label extractor
                    splitter=RandomSplitter(valid_pct=0.2, seed=random.randint(0, 10)),
                    item_tfms=RandomResizedCrop(224, min_scale=0.5),
                    batch_tfms=aug_transforms() )
  

3. Adjust Visualization Size #

Sometimes plots render but are too small to notice. Try specifying a larger figsize:

dls.show_batch(figsize=(12, 12))
learn.show_results(figsize=(12, 12))

4. Validate Your Data Loader #

Make sure your data loader is actually loading images correctly. Print a sample from your dataset to confirm:

print(dls.train_ds[0])  # Should output an image tensor and its corresponding label

If this returns valid data, your issue is definitely with visualization rendering, not data loading.

5. Check Matplotlib Version Compatibility #

FastAI v2.3.1 works best with specific Matplotlib versions. Outdated or overly new versions can cause rendering bugs. Pin or upgrade to a compatible version:

pip install --upgrade matplotlib==3.5.3  # A version known to play well with FastAI v2.3.1

内容的提问来源于stack exchange，提问作者Yunus Gedik