jwtrash67

Second devlog - This took some time.
Since I want my program to be able to detect faces and license plates reliably, I choose the process of finetuning a YOLO model. This also gives the advantage of existing implementations for the coral USB accelerator. But the biggest issue was to find datasets to pretrain. I found a similar project using a general use dataset, but I wanted to use specific ones, since I hope that this will result in an improved accuracy. So I’ve settled on the Large-License-Plate-Detection-Dataset on Kaggle and the CUHK-CSE/wider_face dataset. I downloaded them using a python script, and wrote a small script to transform the CUHK-CSE/wider_face to the necessary YOLO structure. Now I will research on how to best use both datasets. Do I need to join them together? Or do I train the model consecutively or parallel with them? I don’t know, and I will try to find out and also do exactly that

First devlog - getting the Coral USB Accelerator up and running on my Windows machine

After installing the drivers, setting up a virtual environment and installing all necessary libraries, I’ve adjusted a demo script to run under windows and as you can see, IT WORKS!!