Share your work here ✅

Thx. If u have any questions or ideas how to improve it let me know :slight_smile:

Sounds like your data collection/curation approach might be worth a blog post!

I don’t know if it’s really the right place but can you guys please like this issue? Fast.ai Integration · Issue #2941 · mindsdb/mindsdb · GitHub.

There is a competition for best integration idea and the MindsDB library is super useful.

I have created Marvel Heroes Classifier. Although it cannot guess much correctly. The model can guess correctly when the photo contains distinct pattern (Shield in Captain America, Arrows and bow in Hawkeye, etc).
I trained for about 6 epoch.

Also created the simple website using api from huggingface. Additionally while guessing it extract related Wikipedia page data.
http://marvelheroes.soepaing.com

2 Likes

Hi there, are you all right?

I have a big interest in AI generated art, so I made a classifier based on the one from Lesson 1 (Is it a bird?) that classifies if a AI generated picture is generated by dalle2 or by midjourney. Here’s the link to the notebook, hope you enjoy!

2 Likes

I’ve deployed a bird species classifier for over 600 species living in France.

It’s got a decent accuracy of about 90%, with most mistakes on species from the same family that can be notoriously hard to distinguish.

I wrote a blog post describing how I approached the task and sharing my thoughts/appreciation for the fastai course/book combo.

5 Likes

I created a classifier to identify the brand of different luxury handbags. It gets to ~85% accuracy when identifying among 5 different brands. Feel free to try it on huggingface!

Fun fact: I first tried to identify sunsets from sunrises, but apparently they are non-distinguishable. The model had only 55% accuracy, barely better than random guessing

2 Likes

Our paper on finding ancient Maya settlements in LiDAR with Deep Learning (and winning the 2nd prize in the #ECMLPKDD European Conference on Machine Learning 2021 Discovery Challenge) is finally online:

Key ingredients: Fast.ai and data augmentation via synthetic data.

5 Likes

New Zealand bird classifier now on Hugging Face. Yay!

Also discovered someone beat me to the it by a few years: Share your work here ✅ - #1695 by pete88b

2 Likes

Hello Everyone,

Now a days, In India “Cheetah” has been making News. Almost 70 years ago, “Cheetah” has nearly wiped out in India. But, Now Indian Government has brought 8 “Cheetahs” from South Africa.

I was watching news in T.V. , It was shown that, one of the reason that “Cheetah” has become vulnerable animals in the world is that people hunt it, misunderstanding it with " Leopard" which is far more dangerous than “Cheetah” but they kind of look same. So, I thought of making a model to classify them.
I have completed my two lessons of the book and really having fun of making projects. So, just one to share one of the project with you all.

I write a Blog about this. I hope you all enjoy.

kaggle link-

6 Likes

Completed lesson 1 of the course and have been thinking of cost effective ways that could prevent major fire accidents at all locations. With CCTVs widely being used everywhere, I believe that the hazardous work locations (including kitchen at home) could be monitored using the image classifier for smoke and fire, and immediately detect and raise alerts to prevent major fire accidents. Here’s the notebook[1] that helps to detect if the given CCTV footage has fire or smoke in it.

[1] - fire or smoke - fastai | Kaggle

1 Like

Wow, that sounds great! However, it seems that the link doesn’t work. (At least, for me.)

Thanks for the notification, fixed it!