School of AI Beijing Study Group

Hi,
I am trying to create my data set from google images. I have my txt file, but it cannot be found when I want to download it.

could you post the output of path.ls() ?

For lesson 2 exercises, I’ve following errors
Anyone could help where might be wrong?
thanks

Never mind. The error just disppears after I rerun all together several times… I didnot change any code…

No worries, I found it the problem, actually I was not using the right path

If encountering the same issue with colab as I did :laughing:, please see below two posts


In short, colab does not support FileDeleter widget or the likes, such as ImageCleaner, and FileDeleter is removed in favor of two new widgets: ImageDeleter and ImageRelabeler .

And there is a work around below

Since the original lecture video, the fastai library was updated. Most likely explanation is that FileDeleter was removed and instead you should try using ImageDeleter that you mentioned.

One of the thread says it’s the security model of colab does not support ipywidgets, and I tried ImageDeleter still reports “not defined” error.

Hello guys!
Am trying to clean my data. But each time i try to run ImageCleaner, i encounter the same problem all over again, hosted runtime is “BUSY”. I restarted more than once all runtine but still having the same issue. Any solutions?
Thanks!

See above post links, ImageCleaner is not supported in Colab.

1 Like


Hello there,
I tried to test images that have nothing to do with bears after deploying the classifier app. And however the image is classified into one of the classes of my model. Is this because the error_rate? Maybe this is because there is no instructions for if the images couldn’t be classified ?

Hi @maxim.pechyonkin

Thanks for this Beijing study group meetup! I am sorry to miss out on the April one, will there be more meetups?

Our weekly meetups are held each Monday from 7 to 9 pm in UIBE’s 求真楼, first floor at the international study center. You are welcome to join!

If you train the model with only 3 classes of bears then it will only predict those classes of bear, even if the image is completely irrelevant. Deep learning model is very rigid in this sense - it will only predict the classes that you used during training.

1 Like

Hello,

Is this group still going through the course?

I will be in Beijing for work in June and looking for a like-minded study group. I’ve introduced lots of people to the fast.ai approach back in North Carolina through a deep learning meetup group.

I see people here have used colab and Crestle, but I’m curious if I will still be able to connect to my Paperspace instances. Anybody have experience with that?

Oh, I found the website. Great!

Welcome to Beijing! We have meetings on Mondays. Let us know when you are in Beijing and we can arrange to meet. :handshake:

Sounds great! I’ll reach out and see you in mid-June.

It took me sometime to find it out, so I’ll reply this old question:

First FileDeleter have changed to ImageDeleter and ImageRelabeler and then both have been merged to ImageCleaner which requires an additional parameter ‘path’. For usage one can get instructions with “help(ImageCleaner” at jupyter notebook.

It worked here for me with:
from fastai.widgets import *
ImageCleaner(ds, idxs, path_img, 5)

5 = number of images that will appear each time.

Bests.

2 Likes