So let say you want to training a neural networks to identify something in image. The first step is not actually code it. The first step is data gathering. The next step is labeling. You may also want to identify multiple objects in your image, if so you also need to label the items in the image.
What we do in the case for object detection is, we essentially draw a square or a region and set a name.
[Read More]
Computer Vision The Boring Part - Open Image Dataset
We have shown how one can create your own https://devkami.com/blogs/2021/01/11-ml-labeling/. Another way to get dataset is the Google Open Image Dataset.
This is a dataset that have 1.5 million annotations for 600 categories. It covers some common thing that trained in YOLO. You can have a shot at trying to see if the category exist.
To use this, I suggest that you use a tools, the one I use is the OID Toolkit, I use the fork by The AI Guy a youtuber talking about these.
[Read More]
Avatarify
Recently I discovered a thing called First Order Motion Model. This is a technique to create deepfakes using a driver video and one image. It have a few interesting use, other than deepfake. One particular one is testing clothes.
But today I am doing something more mischievous. I am doing this to pretend to be someone on webcam in a chat. The way I do this is Avatarify. This is an application running on a computer that create a virtual webcam for you to pretend as someone else.
[Read More]
DevKami 116 - Web Hardware
Enchanted Objects - Review
I got this book not too long ago. The book is about design of common things enhanced by embedded devices, or common called in our circle IoT device. This is a more high level book than a technical one, but it have very good insights on how to make an IoT devices.
The book only covers briefly on tablets and devices with screens. The focus is on common device enhanced by embedded computers, or as author says Enchanted Objects.
[Read More]
Happy 2021
Happy 2021. Just a greetings to remind people that this blog exist. No much changes from us this year, which really means we will still wing it most of the time. Experiment ourway out in our content.
We try to do more interview this year. More content to the blog. Finally have tha patreon running etc. Otherwise mostly the same
HAPPY 2021