Daily Newsletter — 24th December 2020
“Christmassy” images with GAN, Neural Network Data Glove with TF and Re-photography with StyleGAN2 in today’s Data Science Daily 📰
Training a Christmas GAN 🎄
Hasty.ai is here with an idea, one that feeds a GAN with a random picture and then outputs a “Christmassy” version. Check out their blog post with the link below to find out how they achieved this using General Adversarial Networks.
The main learning was that it’s not easy to train a GAN to translate images based on an abstract mapping concept like Christmas, instead of specific entities like stripes on a Zebra. Data has a significant impact in these scenarios, as the training formulation is loosely constrained, data is unpaired, and there is little to no supervision.
With minimal effort, they were able to achieve more than decent results and also get a clear image on the working of GANs through this “christmassy” side project.
Read the entire post here 👉 https://medium.com/hasty-ai/building-a-xmas-gan-f4d809a3d88e
Somatic: Neural Network Data Glove 🧤
The Somatic is a wearable keyboard and mouse. It translates hand signs and motions into actions, like the somatic component of a spell in Dungeons and Dragons.
Each knuckle has a Hall sensor, and the first segment of each finger has a magnet. Flexing a finger pivots its magnet out of position, allowing the Somatic to map your hand. An EM7180SFP IMU near the thumb provides 9-degree tracking. Eventually, this will allow you to move a mouse cursor by pointing, and type letters by drawing them in midair.
Project Page: https://hackaday.io/project/169297-somatic-wearable-control-anywhere
YouTube: https://youtu.be/6raRftH9yxM
Assembly Instructions: https://www.instructables.com/Somatic-Data-Glove-for-the-Real-World
Design files, training utility, and pre-trained network: https://github.com/ZackFreedman/Somatic
The Somatic project is a work-in-progress. The objective is to acquire samples of each letter, so a TensorFlow Lite model running locally on the glove can detect handwriting. This is still a long way off, but you can install the provided firmware to stream data back to a computer over Bluetooth or USB.
The training utility is fully functional and can rapidly acquire plenty of test gestures. The feature to train a neural network against the collected data, and transmit the model to the glove is currently in the works.
Machine Learning on the Edge: https://youtu.be/iTj0lcVSIVU
The Somatic project’s priorities are:
- Control any wearable computer with a heads-up display
- Ready to use all day, instantly, with no Internet
- Doesn’t cause fatigue or interfere with other tasks
- Fast enough to do a quick search in less than 10 seconds
Time-Travel Re-photography with StyleGAN2 📷
Many historical people are captured only in old, faded, black and white photos, that have been distorted by the limitations of early cameras and the passage of time.
This paper simulates travelling back in time with a modern camera to rephotograph famous subjects. Unlike conventional image restoration filters which apply independent operations like denoising, colourization, and superresolution, this project leverages the StyleGAN2 framework to project old photos into the space of modern high-resolution photos, achieving all of these effects in a unified framework.
A unique challenge with this approach is capturing the identity and pose of the photo’s subject and not the many artefacts in low-quality antique photos. On comparison to current state-of-the-art restoration filters, this model shows significant improvements and compelling results for a variety of important historical people.
Project Page: https://time-travel-rephotography.github.io/
YouTube: https://youtu.be/eNOGqNCbcV8
Paper: https://arxiv.org/abs/2012.12261
Contact Us 📞
Reach out to us on community@jovian.ai to get featured here. Learn data science and machine learning with free hands-on data science courses on Jovian.
Follow us on Twitter, LinkedIn, and YouTube to stay updated.
