ImageClassifierCleaner not showing images (kaggle)

imju · March 22, 2023, 5:28pm

I am using Kaggle notebook.
The ImageClassifierCleaner(learner) is not showing any images for any label while there some losses.
Is there any chance to bring up the images that were lost?

from fastai.vision.widgets import *
cleaner = ImageClassifierCleaner(learn_resnet50)
cleaner

kurianbenoy · March 22, 2023, 6:33pm

Are you using it kaggle? Can you share your notebook in public so we can also help you to debug this issue.

imju · March 22, 2023, 6:47pm

Thanks @kurianbenoy!
Here is the link.

kurianbenoy · March 22, 2023, 7:43pm

No images coming are a bit strange.

imju · March 22, 2023, 9:35pm

Right. That’s where I am stuck at

galopy · March 23, 2023, 1:08am

Hi,

I’m not 100% sure why, but you could try to download images using duckduckgo and then use the cleaner.

I think the cleaner’s purpose is to delete mislabeled or wrong images.

imju · March 23, 2023, 5:10pm

The image files are part of the well known dataset, Oxford flower 102 types Visual Geometry Group - University of Oxford
Are you alluding that because of that, this ImageClassifierCleaner would not display?
Yes, I have seen this API working with the results from DuckDuckGo or Bing.
Let me dig into the code to see what is going on.

kurianbenoy · March 24, 2023, 2:37am

Yeah I don’t think it’s not because of the dataset the issue happens. In PETs dataset, when I experimented ImageClassifierLearner was working. So it obviously working elsewhere as well in Duckduckgo or bing API.

@imju I think you are going in the right direction, if you are looking at source code of ImageClassifierCleaner. I feel it might be a bug in fastai library. I have shared the code for it.

github.com

fastai/fastai/blob/master/fastai/vision/widgets.py#L105-L123


      
          @delegates(ImagesCleaner)
          class ImageClassifierCleaner(GetAttr):
              "A widget that provides an `ImagesCleaner` for a CNN `Learner`"
              def __init__(self, learn, **kwargs):
                  vocab = learn.dls.vocab
                  self.default = self.iw = ImagesCleaner(vocab, **kwargs)
                  self.dd_cats = Dropdown(options=vocab)
                  self.dd_ds   = Dropdown(options=('Train','Valid'))
                  self.iwis = _get_iw_info(learn,0),_get_iw_info(learn,1)
                  self.dd_ds.observe(self.on_change_ds, 'value')
                  self.dd_cats.observe(self.on_change_ds, 'value')
                  self.on_change_ds()
                  self.widget = VBox([self.dd_cats, self.dd_ds, self.iw.widget])
          
          
    def _ipython_display_(self): display(self.widget)
              def on_change_ds(self,change=None):
                  "Toggle between training validation set view"
                  info = L(o for o in self.iwis[self.dd_ds.index] if o[1]==self.dd_cats.value)
                  self.iw.set_fns(info.sorted(2, reverse=True).itemgot(0))

Do let us know if you are able to solve this issue. Hopefully by next week, I can look in more detail to this issue.

aaron.jeffery · March 24, 2023, 9:23am

If you’re having issues with ImageClassifierCleaner not showing images on Kaggle, there are several things you can try to resolve the issue:

**Check the File Paths:**

Make sure that the file paths specified in your code are correct and that they are pointing to the correct directory where your images are located. Double-check that the file paths are relative to the root folder of your Kaggle project.

**Check the Image Formats**:

Make sure that your images are in the correct format and that they are compatible with ImageClassifierCleaner. The most commonly used image formats for this purpose are JPEG, PNG, and BMP. Ensure that the images are not corrupt or damaged and that they can be opened using a standard image viewer.

**Check the Image Size:**

Check that the images you’re trying to load are not too large or too small for ImageClassifierCleaner to handle. If the images are too large, it may take longer for them to load, and if they are too small, they may not be visible at all. Try resizing your images to a more appropriate size.

**Check the Code:**

Double-check your code to make sure that you’re using ImageClassifierCleaner correctly. Ensure that you’re importing the necessary modules and that you’re using the correct syntax. Refer to the official Kaggle documentation for guidance on how to use ImageClassifierCleaner.

**Clear Browser Cache:**

Clear your browser cache and cookies, then reload the page. Sometimes, cached files can interfere with the proper functioning of ImageClassifierCleaner, and clearing the cache can help resolve the issue.

**Try a Different Browser:**

Try using a different browser to see if the issue persists. Sometimes, certain browsers may have compatibility issues with ImageClassifierCleaner, and using a different browser can help resolve the issue.

Conclusion:

Overall, if ImageClassifierCleaner is not showing images on Kaggle, there are several potential causes and solutions to consider. By checking file paths, image formats, image sizes, code syntax, clearing your browser cache, and trying a different browser, you can increase your chances of successfully using ImageClassifierCleaner to label and organize your image data.

regards windows 11 update: Windows 11 Update | Free Upgrade PCs to Microsoft Windows 11 OS | Lenovo US