Fastai2 Blog Posts, Projects, and Tutorials

[yeldarb (Brad)]

Personal Blogs:

For getting started with a blog, I recommend FastPages.

@arora_aman fastaiexplained:

• DataBlocks API - A high level introduction with examples
• MultiLabel Classification - Free Sound Audio and ATLAS Protein Classification (using four channel images)

@DrHB

• fastai JIT

@Epoching Epoching-Blog:

• Semi-Supervised with FastAI

@Albertotono Design and Deep Learning, discussing about applications in Geometrical Deep Learning in the Architecture, Engineering and Construction space.

@muellerzr fastblog

• fastai and New DataBlock API
• The Idea of a Transform

@clck10 - Looking into fine_tune and general projects

• First part on learn.freeze
• Sanity checking fastai v2 data and models
• Complex-valued Parameter Initialization - Part 1

@barnacl

• DataBlock-Playground, play with small datasets and set up DataBlocks

@harish3110’s Through Tinted Lenses

• Relearning to learn, deep.
• Building an image classifier using Fastai V2
• A look at Fastai’s ‘L’ data structure
• Improving upon flowers-102 baseline model

@farid

• 3 ways to pip install a package: fastai2 use-case

@akashpalrecha

• Using separate Transforms for your training and validation sets in fastai2 : goes deep into splits and split_idx usage inside of fastai2’s low-level Datablock API

@amritv

• Medical Imaging using fastai : Reviewing fastai 60_medical_imaging notebook
• Starting with Medical Imaging Starter notebook that looks at some high level considerations when evaluating model results such as PPV, NPV, Specificity, Sensitivity
• Getting to know DICOMS : Taking a deeper look at how to view and manipulate DICOM images
• Image Augmentation

@hallvagi dl-explorer
Doing a deep dive of ULMFiT for a Norwegian language classification task.

@morgan ntentional blog

• Explore and clean your text data with UMAP clustering
• ICLR 2020 - Efficient NLP Transformer highlights
• ICLR 2020 - Efficient Deep Learning highlights

@jwuphysics Astrophysics research blog

• Deep learning with galaxy images - my first blog post!
• Training a deep CNN to learn about galaxies in 15 minutes - a technical post about training CNNs for regression in astrophysics
• Visualizing CNNs, part 1 - exploring Pytorch/Fastai Hooks for visualizing CNN activations
• Visualizing CNNs, part 2 - implementing and understanding Grad-CAM for a galaxy classification problem

tyoc213:

• Reconociendo dígitos en fastai2 -first blog post on how to use fastai2 for kaggle digits competition.

karynaur

• Classifying cats and dogs using fast.ai’s vision library-First blog on image classification
• MNIST Handwritten digits classification from scratch using Python Numpy - A beginner friendly blog on writing a deep learning model to classify MNIST in vanilla python.

Hannes (@ johannesstutz) Deep Learning Berlin

• Understanding fastai’s Midlevel API How to customize your data processing. Exploring Transforms, Pipelines, TfmdLists, Datasets

Blogs with General advice

• How to do machine learning efficiently
• Machine Learning and Testing
• How not to do fast.ai (or Any ML MOOC)

• Fast.AI
• Another data science student @sgugger

Packages

@rsomani95:

• fastai2_extensions: Utilities for interpretation, exporting, and some augmentation. Link to forum post

Youtube Videos:

• Interviews fastai alumni

Projects:

• Official Projects Channel:

• @muellerzr

  • A walk with fastai2
  • FastGarden

• @giacomov Not all datapoints are equal

• @mgh COVID19

• @Albertotono COVID19 2nd

• @tcapelle AutoEncoder

• @lgvaz

  • faststyle
  • fastcook

• @amritv

  • visual_ui
  • fastgui
  • Medical Imaging

[muellerzr (Zachary Mueller)]

Absolutely. I’ve been doing my Walk with fastai2 study group for a few months now, just so long as there is no direct course stuff you should be fine

[Albertotono]

Dear @yeldarb I would like to share in my Blog also my journey with fastai. www.bimhox.com
Let me know if there are specific guidelines to follow?!
Best Regards

[jeremy (Jeremy Howard)]

@Albertotono

Let me know if there are specific guidelines to follow

Just don’t share anything from the course itself - or if you do, use a private link. But feel free to write your own post that states things in your own words, and do mention that you’re learning from a fast.ai course that will be publicly available in July.

Thanks for asking!

[Albertotono]

@muellerzr @yeldarb @jeremy can we make this post a wiki, so people can here their fastpages blog, or other project, blogs, youtube channels etc in this unique post?

[yeldarb (Brad)]

Good idea, see if that worked.

[muellerzr (Zachary Mueller)]

Awesome idea! And I moved walk with fastai2 as a project, I plan on doing a seperate blog series (and as it’s not quite a blog per say). Though as I say that… you do have me thinking. Lemme try something

[Albertotono]

Oh awesome, sorry for the mistake Zach, keep us posted.

[muellerzr (Zachary Mueller)]

@Albertotono

sorry for the mistake Zach

All good Well, the idea was me considering how easy it would be to convert over those notebooks into blog form, but thinking about it I would need some substantial rewrites so I’ll just leave it as a project

[Albertotono]

or maybe a book

[muellerzr (Zachary Mueller)]

@Albertotono I migrated it over here. As folks will probably be writing many many blog posts, I think the general recommendation is if you’re posting multiple blogs, put them under a branch in your username so it has some structure (Feel free to edit/format differently if you’re not a fan of it!)

[tcapelle (Thomas)]

Hello!
I put a small example on how to construct an autoencoder using V2.
This is a Image2Image model using the DataBlockAPI
It runs on colab, you can view it here
I struggled a little bit, so it may be useful for someone else.

[clck10 (Chris Kroenke)]

Added the first in a (hopefully) series that does a deep dive on fine_tune. also my excuse to try fastpages, loving it so far!

[clck10 (Chris Kroenke)]

Added a second post on some visualizations and sanity checks that have helped me in v2 so far. It is a first-pass list, and I would love to grow it with suggestions from others!

[muellerzr (Zachary Mueller)]

@clck10 I didn’t see datablock.summary() in there, this is something you should investigate carefully and would be a great addition to the article. I use it always before I even build a DataLoader

(Also removes much of the boilerplate code for testing the DataLoader you had )

[clck10 (Chris Kroenke)]

Hey Zach! Thank you. Very silly question, but how do I get a data block from the data loaders? I had a datablock.summary() in another project where I used that API, but stripped it away for this example that goes straight to dataloaders.

[muellerzr (Zachary Mueller)]

Not silly! I had forgotten you were looking solely at the mid-level (and very high level) API and so those are excellent debugging tools I had brought it up because it looks very similar to how the high level datablock.summary() looked! So you’re on the perfect track

[clck10 (Chris Kroenke)]

Ah gotcha, thanks! You are right though, the datablock.summary() is incredibly useful. Plus writing these with fastpages is actually enjoyable! First time where the publishing medium has made me want to keep going.

[muellerzr (Zachary Mueller)]

@clck10

First time where the publishing medium has made me want to keep going.

It really is such a nice setup and so easy to do. Can’t wait to read more from you

[DrHB (Asimo)]

Hi all =)
My first blog post, please add to wiki =)

fastai models to torch.jit

fastai models to torch.jit

This page is really helpful to keep track and cite people work! so we don’t repeat stuff. I have already cited @muellerzr and @arora_aman =)