#105 Colorizing and Restoring Old Images with Deep Learning
Python Bytes - A podcast by Michael Kennedy and Brian Okken - Luni
Categories:
Sponsored by DigitalOcean: pythonbytes.fm/digitalocean
- Text interview by Charlie Harrington of Jason Antic, developer of DeOldify
- A whole bunch of machine learning buzzwords that I don’t understand in the slightest combine to make a really cool to to make B&W photos look freaking amazing.
- “This is a deep learning based model. More specifically, what I've done is combined the following approaches:
- Self-Attention Generative Adversarial Network
- Training structure inspired by (but not the same as) Progressive Growing of GANs.
- Two Time-Scale Update Rule.
- Generator Loss is two parts: One is a basic Perceptual Loss (or Feature Loss) based on VGG16. The second is the loss score from the critic.”
- via Jason Pecor
- PlatformIO is an open source ecosystem for IoT development
- Cross-platform IDE and unified debugger. Remote unit testing and firmware updates
- Built on Visual Studio Code which has a nice extension for Python
- PlatformIO, combined with the features of VSCode provides some great improvements for project development over the standard Arduino IDE for Arduino-compatible microcontroller based solutions.
- Some of these features are paid, but it’s a reasonable price
- With Python becoming more popular for microcontroller design, as well, this might be a very nice option for designers.
- And for Jason’s, specifically, it provides a single environment that can eventually be configured to handle doing the embedded code design, associated Python supporting tools mods, and HDL development.
- The PlatformIO Core written in Python. Python 2.7 (hiss…)
- Jason’s test drive video from Tuesday: Test Driving PlatformIO IDE for VSCode
- Nice overview of visualization landscape, by Anaconda team
- Differentiating factors, API types, and emerging trends
- Related: Drawing Data with Flask and matplotlib
- Finally! A really simple example app in Flask that shows how to both generate and display matplotlib plots.
- I was looking for something like this about a year ago and didn’t find it.
- Full Visual Studio Code, but in your browser
- Code in the browser
- Access up to 96 cores
- VS Code + extensions, so all the languages and features
- Collaborate in real time, think google docs
- Access linux from any OS
- Note: They sponsored an episode of Talk Python To Me, but this is not an ad here...
- Forbes’s article about Guido and the Python community actively working to get more women involved in core development as well as speaking at conferences.
- Good lessons for other projects, and work teams, about how you cannot just passively “let people join”, you need to work to make it happen.
- From Anna-Lena Popkes
- Plain python implementations of basic machine learning algorithms
- Repository contains implementations of basic machine learning algorithms in plain Python (modern Python, yay!)
- All algorithms are implemented from scratch without using additional machine learning libraries.
- Goal is to provide a basic understanding of the algorithms and their underlying structure, not to provide the most efficient implementations.
- Most of the algorithms
- Anna-Lena was on Talk Python on 186: http://talkpython.fm/186
Extras:
- Michael: PSF Fellow Nominations are open
- Michael: Shiboken has no meaning
- Brian: Python 3.7 runtime now available in AWS Lambda