Lesson 2: further discussion ✅

(Paul Baclace) #147

It would be nice if there was an option to remove the actual image; sometimes one wants to use the image collection for another purpose without having to use the same interface that requires cleaned.csv to tag along.


Bump! Yea I’m having this issue too… ImageCleaner just hangs, if I remove the path argument I get the same error. Did you find any way around this?


Hi there! I’m testing out with my own data and ran into error while trying to put predict a single image.

Seems like learn.predict is returning the correct result but pred_class is displayed as a memory address rather than a label, any idea why?


Nvm I figured it out, just needed to call print(pred_class)


Hi I am trying to train a classifier for medical images but learn.fit_one_cycle(4) does not seem to be doing any training.


I found someone posted saying Colab didn’t support widget ImageCleaner ( ImageCleaner ). There’s an alternative which is little tedious, but it does solve the problem. I used this method and able to get the dataset cleaned and proceed the subsequent steps. Here’s the alternative solution I found in other post : ImageCleaner missing argument in lesson 2 download notebook


We’d probably need to have a look at how you created the databunch to guess what’s going on

Edit: by the look of it, it seems ALL your images are labeled as “Key_slice_examples”. So the model is always predicting that one single label, with 100% accuracy, and no loss


Did you manage to figure this out?

(Varun Verma) #155

So I was creating my own classifier on distinguishing between spongebob and spiderman and


does not output and anything and the graph is empty too. I am doing on this on a kaggle kernel if that provides any helpful info. Anyone encountered a similar problem?

(Azwar Khalid) #156

@joaoeira The method declaration is wrong, remove the last parameter Path as it is not required. it should be:
ImageCleaner(ds, idxs)

Type help(ImageCleaner) to see the details

(Azwar Khalid) #157

Also I realised it was an issue with a version of ImageCleaner. The older version did not have the path parameter the newer version has path parameter. So you will need to upgrade fastAI. Do ??ImageCleaner to see which version you have

(Butch Landingin ) #158

Hi all,
I created a Github project template for deploying simple image classifiers on Render.com
based on render examples as described in the deployment docs.


Hi there. Unfortunately, I did not figure this out. I could not reproduce the particular way that the library calculates its derivatives.