1. Creating dataset - abhishekkumardwivedi/android-things-AI-Camera GitHub Wiki

VOC dataset

We can either use ready available datasets or we can build our own dataset for the specific use case. These details guides building our own dataset. We will be building VOC dataset by using ImagNet utility. VOC basically provides a format in which we are building the dataset. To create dataset, we need a large number of images, in multiple of 1000s. Since it takes literally huge horsepowers to process all the images, Here I am using only about hundred images to show the demo. As we are looking for a specific class of an object and that too very precise case, one kind of number plate, right from the front view. This is to let us verify our 1st model in less of time. Basically, we are taking a lot of images and highlighting the object within XY plane and providing object labeling.

Annotation with labelImg

I used labelImg and it became my favorite soon. labelImg is python based tool.
Following are the steps to install labelImg.

$ sudo apt-get install pyqt5-dev-tools
$ sudo pip3 install lxml
$ git clone https://github.com/tzutalin/labelImg.git
$ cd labelImg
$ make qt5py3
$ python3 labelImg.py
// set default annotation directory and label in beginning to speedup 

Other tools for image labeling

There are tools, such as BBox , which could also be used for image labeling.