Part 2 focuses on recognizing edges, and curves of complex shapes.
Have fun! 
Maria
2 Likes
This blog elaborates on the CNN ins and outs - channels, size changes, activation function and batchnorms. En couleur!
Enjoy!
Maria
2 Likes
Thank you for this post. I’m currently on lesson 6 and needed to collect datasets from Kaggle to Colab
I’m three lessons in, and I trained a model to detect redacted pages for FOIA (freedom of information act) requests. I achieved around 96-97% accuracy, which isn’t perfect, but I’m pretty happy with where I got so far! This was the kind of image I was trying to be able to classify (from public sources)
I wrote up my process of labelling the data and some of the tradeoffs I could reason through for the model here: Training a classifier to detect redacted documents with fastai | mlops.systems
3 Likes
Object Detection using the Fastai library embedded within the IceVision framework.
Enjoy!
Maria
2 Likes
Get acquainted with state-of-the-art models used for object detection. The IceVision framework utilizes the Fast.ai library, so the syntax and structure should already be familiar to you
Have fun coding!
Maria
Can a machine learning model identify an ill baby?
My attempt can be viewed at:
Thank you everyone who made a contribution on this journey!
7 Likes
For someone who want to practicing customizing fastai on a new task, I’ve just create a notebook here on Kaggle RSNA fastai DataLoader | Kaggle . It is about creating a fastai DataLoader for sequence of Images (MRI Images, Video, …).
Hope it helps,
3 Likes
Continuing on to object detection using the Fast.ai library in an IceVision framework.
Maria
3 Likes
Hi!
I wrote a tutorial about training a MnasNet 1.0 image classifier using FastAi, converting it to ORT, and deploy to an Android app. Really simple to do! 
I hope you enjoy it! 
5 Likes
Great work! I learned Kotlin over the summer so super anxious to test out my skills
1 Like
For the past couple of months I experimented 
with weaving together @fastdotai, TIMM by @wightmanr and nnAudio for a custom task as part of the @kaggle G2Net competition. I have finally put together a “How-To” blog post explaining my approach on my new website.
The blog is available here 
Extending Fastai For Custom Task - Satyabrata pal
P.S. – My other works using FastAI are still in my old blog and can be read here-- ML and Automation – Medium . I am planning to gradually move them to my new blog.
1 Like
As a beginner-level project, I used the American Sign Language (ASL) dataset from Kaggle to predict the letters of the sign language given an image.
I was able to predict all of the test images in the provided data set accurately. However, noting that the test set only consisted of 28 images, I looked to testing my model on custom images as well.
I have written a blog on my initial experience with fast ai, filled with external resources for help as well. I hope to increase my work’s reach and learn from this incredible resource.
Here is a link to my blogpost:
Thank you in advance for reading and sharing my work.
My budding twitter profile is: https://twitter.com/qasimr97
2 Likes
Detecting leaf disease using Fastai/ IceVision frameworks.
Maria
4 Likes
I’ve just created a notebook for migrating Pytorch model to fastai for ventillator pressure competition in Kaggle Ventillator Pressure / Fastai | Kaggle
Many things in pytorch can turn into fastai without much works I think.
Hope it’s useful
2 Likes
The blood components are unbalanced - we have predominantly more red blood cells compared to the other types. But the visual characteristics of the 3 types studied (red blood cells, white blood cells and platelets) are sufficiently distinct so that the model is able to perform well despite the imbalance.
Have fun coding!
Maria
3 Likes
I’m trying to explain some mathematical concepts in an easier manner on twitter. So if anyone is interested in, you can take a look at: https://twitter.com/DienhoaT
- Classification Loss Function: https://twitter.com/DienhoaT/status/1454551383277981702
- Mean Squared Error vs Mean Absolute Error: https://twitter.com/DienhoaT/status/1454203557968760833
Some more concepts will come soon.
Thanks
I’ve just written a starter notebook for tabular playground competition in kaggle, just used the default config of tabular model. tabular playgroud - fastai | Kaggle . If someone is interested on, please fork and see if we can improve it. Thanks
1 Like
I too recently gained some proficiency with the Tabular side of fastai. For a Hackathon on Saturday centered around music industry data, I made a Colab notebook for predicting song popularity using some features obtained via the Spotify API. And I worked in some model interpretability tools and visualization of embeddings:
3 Likes