after going through the first video and chapter 1, I tried building an Indian Car Image Classification model based on the body style. Here is the Kaggle Notebook for the same.
Learnings from Chapters 1 and 2:
The model has been trained on Bird images and forest images. So it will be effective in distinguishing a bird from a forest.
Note: I did not create a Bear classifier. If you upload bear photos or any other animals you might find surprising results.
Blogpost:
https://mdkhalid-siddiquee.medium.com/model-deployment-birdclassifier-7e383016c1e0
Link to deployed model:
Ml Production - a Hugging Face Space by mdkhalid
Link to notebook to create the classifier model:
Is it a bird? Creating a model from your own data | Kaggle
Different style of blogging: I have embedded google slides in Medium to directly present my PowerPoint rather than creating a blog. I hope you find it useful.
Rubanza, this is amazing work! I read your blog post and it was really nice how you show every step. I also love your small “sidebars” with your own thoughts. Keep up the great work!!!
1 Like
Thanks for the read and thumbs up.
1 Like
Me & three of my colleagues managed to win a silver medal in the “Benetech - Making Graphs Accessible” Kaggle competition. We used fastai & IceVision extensively in our solution.
Here’s a blog post on our solution: Kaggle Silver Medal Solution: Making Graphs Accessible – Xelpmoc Data Science
3 Likes
After completing first two demonstration of this course , I thought to (according to my knowledge and interest for anime #weeb) create a classifier that distinguishes from the famous anime characters like Eren Yeager, Levi Ackerman, Tangiro Kamado and Yagami Light.
here’s the link Anime-Character-Classifier
PS: My model classifies Mikasa Ackerman to Levi Ackerman as 63% thus confirming the validity of my model
1 Like
Hi All,
After the first two lessons, I created a simple Ghibli movie detector with 10 classes. It’s got bad accuracy but it’s a start.
Here’s the link to it- https://huggingface.co/spaces/Amulya/Ghibli_movie_classifier
3 Likes
Hi all,
I recently finished a project using ULMFiT from lesson 4. I used the twitter API to scrape twitter replies and then used Fastai to train on the corpus, and then create a text classifier. With ~400 samples of reply text, I was able to create a NLP classifier capable of detecting bot replies on twitter threads with ~95% accuracy.
I wrote a blog about it here: Using NLP to detect scam tweets with Fastai & ULMFiT | by Justin Donnelly | Jul, 2023 | Medium
2 Likes
Just deployed my first model on huggingface Spaces! Thanks to fast.ai’s deep learning course. Exciting times ahead! Cat or Dog - a Hugging Face Space by ashwin3005
1 Like
Interesting work and thanks for the detailed blog!
1 Like
here is the hugging face Hugging Face Test Lay Up Dunk Model - a Hugging Face Space by ali89
1 Like
Dear All,
I am new to the community and haven’t posted a lot here. I have completed watching and tinkering around with first 3 chapters of fastai course part 1. I am a Software Architect/Engineer based in India primarily working on building distributed applications on Azure.
I have gathered courage to pen down my first article on Machine learning as was suggested on the course.
Willing to get feedback on this Unlearning and relearning Deep learning -
1 Like
Hi all,
Just posted my second blog, which is largely based on lessons 3 and 5 from the course. Just wanted to create it to solidify my understanding of some key concepts, as well as include my image classifier for playing cards.
Give it a read if you’re interested, maybe it will help bring some of the concepts together for you. If I got anything wrong, please let me know!
https://medium.com/@justin.donnelly0804/understanding-neural-networks-using-photos-19a670b034a9
2 Likes
Great job!
1 Like
I got inspired by the bird classifier and decided to slightly change it so that it can recognise between a bronthosaurus and a t-rex.
I’m searching the images using DuckDuckGo.
You can find the complete work on my Kaggle account https://www.kaggle.com/andreagrandi/is-it-a-bronthosaurus-or-a-t-rex
1 Like
Hey !
I wanted to share my progress after finishing lesson 2. I decided to train a model to distinguish amphibians from reptiles. I am not a biologist, so it was an interesting learning experience.
I did a few experiments. Kept a log of things I tried (screenshot below 
).
There were cases when I was really surprised. For example I trained the model with the same parameters a few times and was not able to reproduce the best results I got from my first training…
Adding additional batch transforms did not improve results, simple resize was better than resize with crop… My current theory is that my dataset was not perfect.
I hope to understand more, why this was happening after completing a few more lessons.
I published the best model (the first one I trained) on HuggingSpaces: Amphibian or Reptile - HuggingFace Space. Have a go and try it! 
Thanks!
1 Like
Congrats! 
1 Like
Have you ever wanted to know what kind of leaf you’re looking at but were too afraid to ask? Me too, that’s why I trained a leaf classifier model.
The model takes a small but well curated dataset (“UK Leaf Dataset - Original” created by Sameeha Rahman) of 340 images of leaves from 7 categories (~48 items per category). The categories are: maple, nettle, hogweed, blackberry, blackthorn, oak and laurel.
It takes quite a while to train (~40 seconds per epoch and ~ 5 minutes in total even though I’m using a GPU T4 x 2 accelerator on Kaggle), but has ~97% accuracy on the validation set, and is quite accurate on unseen data.
Conclusion
Turns out it’s good at detecting nettles but not much else…
Links
Kaggle notebook that trains the model: https://www.kaggle.com/code/jamesbyers1/leaf-classifier
Dataset: “UK Leaf Dataset - Original” by Sameeha Rahman (link to dataset in my Kaggle notebook. I would link here to give credit but “An error occurred: Sorry, new users can only put one link in a post.”)
1 Like