- End-to-End Style Transfer Tutorial Addendum
- Crop Image on GPU
- Flip Image with Compute Shaders
- Links of the Week
Well I haven’t got around to trying to make daily recaps. I actually didn’t even remember that I was thinking about that until I start writing this post. I made a note on my white board so I don’t forget tomorrow. On the plus side I managed to complete three tutorial posts over the weekend which I think is a record for me.
I completed the follow up post I mentioned in the last recap post covering how to use the video style transfer model that I had been experimenting instead of the model used in the main tutorial. I don’t recommend using the video model over the one from the tutorial but the instructions are there now for anyone interested.
I had worked out how to efficiently crop images in Unity with a GPU while working on my PoseNet tutorial. I didn’t end up using it as I decided to just squish the camera input into a square instead. However, this might not always be ideal so I decided to document how in case I need it in the future.
I also worked out how to flip images using Compute Shaders while working on my PoseNet tutorial. Again, I ended up not using it in the tutorial. I realized after the fact that I could just flip the output of the model instead. This is much less work than flipping the whole image.
I had actually started to write a tutorial for this a while ago but ended up scrapping it. For some reason, Google still picked up the page and someone apparently tried to view it recently. I felt bad that there was nothing but a 404 page waiting for them so I ended up making this over weekend as well.
Links of the Week
This post covers how to implement heatmap regression to perform human pose estimation with the fastai library.
An interesting post exploring how recent neuroscience research may point the way towards defeating adversarial examples and achieving a more resilient, consistent and flexible form of artificial intelligence.
A colossal, cleaned version of Common Crawl’s web crawl corpus.
This blog post covers how to re-implement Stochastic Gradient Descent, Momentum, RMSprop, and the Adam optimizer from scratch.
This GitHub repository contains implementations of popular SOTA self-supervised learning algorithms as Fastai Callbacks.
This Udemy course teaches you how to write Unity Compute Shaders to create particle effects, flocking, fluid simulations, post processing image filters, and create a Physics engine.
This app lets you use your iPhone/iPad to control the virtual camera in Blender, Autodesk, and Maya like an actual camera. You can explore the scene in real-time and record camera motions.
A free setup for an Android-based virtual camera.
An incredible gigapixel mosaic image of the Milky Way.