Defect Detection with Kili AutoML and Weight and Biases

[theodu]

My resource

Topic: An article that demonstrates how to use Kili autoML with Weight and Biases The article will be an image use case: defect detection Among the features of autoML, the article will focus on train.py and predict.py to push annotation. Both autoML and weight and biases will be used and well represented in the article, the articulation and connectivity between both will be particularly emphasized

The mind of the content: an article rather than a detailed tutorial that shows step by step. Main lines of code will still be included to show how to reproduce the key integration features. The idea is the show a proof of value for Kili autoML, how it can be useful, and how to integrate it with famous ML framework from the ecosystem (here Wandb)

Outline:

My content is

• [ ] A Kili Tutorial / Guide / How to article
• [x] An Article

[asmaamirkhan]

Hi @theodu,

Could you please assign this article to me? I'll propose an outline within the following couple of days.

Have a nice day 😇

[theodu]

Great! have a nice day too!

[asmaamirkhan]

Hi @theodu , I prepared the outline like the following, what do you think about it?

Outline

• Introduction
• What is data centric AI?
• What is Kili Technology's AutoML project?
• What is Weights and Biases and how does it work with Kili's AutoML?
• Use case; Surface Defect Detection:
  • Task description and dataset explanation.
  • NEU-DET dataset will be used. To avoid complexity, only scratches class will be shown.
• Traditional pipeline
  • (obtain data ➡ label all data ➡ train)
• Kili Technology AutoML-empowered pipeline
  • (obtain data ➡ label a part of the data ➡ train an intermediate model with the labeled part ➡ use the model to label a part of the unlabeled data ➡ verify the automatically labeled data ➡ train.... )
• Kili Technology AutoML and Weights and Biases empowered pipeline
  • (obtain data ➡ label a part of the data ➡ train an intermediate model with the labeled part ➡ evaluate the model with various metrics ➡ use the model to label a part of the unlabeled data ➡ verify the automatically labeled data ➡ train ➡ evaluate.... )
  • train.py script usage
  • predict.py script usage
  • How to compare models using Weights and Biases dashboard?
• Conclusion

Pipeline explanations will be enhanced with visual diagrams

[theodu]

Hi @asmaamirkhan thank you very much for this outline! it seems very good overall, I will just validate it with other people from the team and I come back to you ;) Have a great day

[asmaamirkhan]

Hi @theodu , Okay! Have a great day too 🚀

[asmaamirkhan]

Hi @theodu ,

Is there any update about the article? 😅

[theodu]

Hi @asmaamirkhan Sorry no update yet from the article, we sent the outline to weight and biases for their validation too and we are still waiting for their answer. Sorry for the delay, I will let you know as soon as we have their answer!

[theodu]

Hello @asmaamirkhan how are you doing ? Great news: W&B also validated your article. You can then begin to write about it. Looking forward to publish this article ! Have a good week, Théo

[asmaamirkhan]

Hi @theodu, Sorry for the late reply. That's great! I'll start to work on it immediately. I'll do my best to finish it as soon as possible but I might be a little bit late because of the exams (not later than 30 Jun).

Have a nice week!

[theodu]

Hi @asmaamirkhan Thank you for the updates! And good luck for your exams Cheers Théo

[theodu]

Hi @asmaamirkhan, How is it going with your article ?

[asmaamirkhan]

Hi @theodu , I'm doing the last touches; I'll deliver it for review before 30 Jun.

Have a nice day!

[asmaamirkhan]

Hi @theodu! The article is ready for review at this link, please take a look!

Note: In the writing guidelines it is mentioned that we have to write the outline of the article after the introduction but I noticed that you are providing an auto generated outline on the website so I didn't write the outline as a part of the article.

Thank you in advance 😇

[asmaamirkhan]

Hi @theodu,

Could you please confirm that you received the article?

Thank you in advance!

[theodu]

Hi @asmaamirkhan, Yes sorry, We confirm that we received your article. We are currently reviewing it and sending it to W&B too Thank you! We will reach you back as soon as we have feedback to give you

[asmaamirkhan]

Hi @theodu,

Waiting for your reply! Thank you for the feedback 😇

[theodu]

Hi @asmaamirkhan We are advancing on the review. Weight and Biases also found your article nice! Could you please share your w&b workspace with us ? W&B would like to make a cool report out of this. And as another comment, we do not see any screenshot of the Kili platform in the blog. Could you please add 1 or 2 ? like showing the queue pages with assets and the labeling interface. Thank you again! Best Théo

[asmaamirkhan]

Hi @theodu, I'm so happy that you and W&B like the article! You can access my workspace using this link. About the new changes; I'm going to add them now, they will be done within a couple of hours. I will drop a new comment here when I'm done.

Have a nice day, Theo! 😇

[asmaamirkhan]

Hi @theodu, I've just finished adding the new parts. I added them under a new section. Please take a look! Also, I wanted to add more screenshots about the labeling dashboard and the analytics section on Kili dashboard, but they're off-topic so I didn't add them to the article. I think that we can shape an outline for a new article/tutorial to introduce the different parts of Kili dashboard. Would you be interested in such article?

By the way, since I'm on my university summer holiday I have time to write another article or tutorial, if you have a specific idea that is waiting for an author, you can share it with me 😇

Best regards ✨

[theodu]

Hi @asmaamirkhan I was off the last week, sorry for the late reply.

• For the image, I wasn't talking about a whole Kili dashboard presentation, but rather one picture of Kili platform with all assets, just to give a concrete overview of the product while people read the article. It is only an illustrations to keep them interested. Could you please add this image ? it would only take 10 sec if you still have the project.
• Thank you very much for the workspace link, I will share it with Weight and Biases! We will have a call with them next Thursday and we will know if we can fully validate the article and publish it! :)
• Great to know that you have free time! We enjoy working with you and we are very happy about your articles! i don't handle the content roadmap but if we think about an article, we will of course reach you to know if you like the topic ;)

[asmaamirkhan]

Hi @theodu, Thank you for your kind reply! Actually, I couldn't understand which page you mean, I haven't deleted the project.

This is the main page of the project and I've already added it to the article. Could you please clarify which page you mean?

[asmaamirkhan]

Hi @theodu ,

How are you? 😇 Is there any update about the article?

Thank you in advance.

[theodu]

hi @asmaamirkhan Sorry I was sick the last week. Thank you very much for the pictures on the article. W&B and Kili validated your article. We are now ready to publish it! Our team will write it on our website and we will send you the link as soon as it is published. You will then be able to send your invoice. Once the article is published on our website, W&B will then publish it on their blog too. Thank you and congratulations for your work again!

[asmaamirkhan]

Hi @theodu ,

I hope you get to feeling better soon! Thank you for your feedback, I'm so glad that they validated the article.

I'm waiting for it to be published impatiently 🚀

Have a great day!

[mtakili]

Hey @asmaamirkhan! Status update: we'll publish your article Tuesday. :)

I'll send you the link once it's live.

Thanks for this top-notch article! We can't wait to share it. ;)

Best, Marianne

[asmaamirkhan]

Hi @mtakili ,

Thank you for the feedback, I'm so happy that it will be published soon! 🥳

Have a great day! 🚀

[mtakili]

Hi @asmaamirkhan! 🙌

Your article is live! its link => https://kili-technology.com/blog/make-training-process-productive-with-kili-automl-and-weights-biases

You can now invoice us for this!

Thanks again for this cool article. :)

[asmaamirkhan]

Hi @mtakili , That's great, I'm so happy about that 🥳! I've just submitted the invoice.

I noticed that you didn't use the cover image that I designed. Did you do that mistakenly? Or is there a reason for that?

[mtakili]

Hey @asmaamirkhan,

Glad you're as happy as we're to have it published. ;)

You're very right. As we've recently changed the type of visuals we use, we couldn't use the visuals you've prepared -sorry for this.

The visual we'll release on social media will still feature your picture and name. Also, if you were wondering: as your name is no longer displayed on the thumbnail, we've added it at the bottom of the article as well.

[asmaamirkhan]

Hi @mtakili ,

I understand, thank you for your detailed reply 😇

Have a nice day 🚀