Updates from the Tinyverse - 5th October 2021
Another week flying by, and our team has been continuously chipping away on multiple fronts. Our engineering team has been busy building our SAAS product Hammer Forge (hint, hint - launching before end of this year). Here is a sneak peak of what the Hammer Forge Studio will be like:
On that note, on to some key updates:
Our co-founder met with all the Hack The North competition winners and had a great time chatting about the future of edge computation, ML, and what the overall industry will look! We’ve also initiated conversations with potential Rune Ambassadors! Our Rune ambassadors will help take Rune to universities and other hackathons!
On a similar front, our team got reached out by Princeton University, asking us if we would be interested in working with them on a hackathon. It was shared that the HackPrinceton team would love to see our team this year. We are very honored to be considered! Stay tuned for further updates on when we are ready for this. We are doubling down our efforts on making Rune the tool of choice for hackathons with edge computing and privacy.
ML and New Runes
Want to build a Computer Vision app for edge computing? Try out our Runes built around top of computer vision!
This week, we have tested our Rune with a large number of Computer Vision Tensor Flow models. Now, we have YOLO-V5, EfficientDet Rune for object detection, DeepLabv3 for Image Segmentation.
If you are not seeing your favorite vision model - drop us a note at email@example.com or ask us on our discord. Even better if you runify a model - you are our ambassador then!
You can find your body orientation using our Pose Estimation Rune and use MiDaS Rune to measure the depth of objects in the image!
It has also become easy to generate an animated image from an image. If you have a low-light image, you can enhance its contrast with MIRNet.
What's next in our Computer Vision segment?
We are planning to support our Rune to be able to draw output predictions on the images. In the future, it will be possible to draw the rectangle, segment, and construct pixel-wise masks around every object and display textual class label plus confidence value on the image.
The core team have been busy working on our continuous integration (CI) and releases system so we can start using our Rust code on mobile devices.
We have also been learning a lot from our experience building new Runes and are trying to incorporate those lessons into our work going forward. A couple of pain points we’ve noticed are that custom proc blocks can be difficult to work with and we need to experiment with a larger range of models and ML frameworks - for example, Rune’s image capability assumes images will be in the form TensorFlow uses, [frames, height, width, channels], while PyTorch uses [frames, channels, height, width].
Our end goal is to support a wide range of hardware platforms but also support a wide range of neural network engine frameworks. The whole point of portable computations!
Q4 2021 will be our busiest time as we are launching multiple production ready products and potentially going live with a few deployments. Stay tuned to learn more, and as always, our team is on a continuing mission to keep building value!
Follow us on our socials: