AI for building identification project
Hello!
I have recently started a project and I would like to know if some people would like to help/participate:
I am trying to train a Neural net to do object (buildings in this case) recognition on any base. The goal is, from a given picture of a base, to correctly identify every single building its level and its position on the base in a way that is not costly for a computer and super fast.
So far, I have been collecting images of buildings from clash-wiki for the first phase. I have 248 categories :
For this first phase, I do not consider the empty/full def as 2 different categories. This will probably come later if I get some success in the first phase of my project.
Here is what my training images look like (They have been created by augmentation from 1 picture of every single building). I am trainning with 20k images atm, I will add some later :
First results :
After 7800 steps, 6 epoch over my 20k images, loss is still slowly decreasing! I have been trainning for 24h+ now on my poor laptop CPU :
I am getting some promising results :
Here are 2 tutorials that inspired me :
