Train a custom object to Tensorflow Object Detection Model

Previous Article – https://wp.me/p6xoZs-3y

To do this there are few steps to follow, there are,

1. Collect a few hundred images that contain your object – The bare minimum would be about 100, ideally more like 500+, but, the more images you have, the more tedious step 2 is…
2. Annotate/label the images, ideally with a program. I personally used LabelImg. This process is basically drawing boxes around your object(s) in an image. The label program automatically will create an XML file that describes the object(s) in the pictures.
3. Split this data into train/test samples
4. Generate TF Records from these splits
5. Setup a .config file for the model of choice (you could train your own from scratch, but we’ll be using transfer learning)
6. Train
7. Export graph from new trained model
8. Detect custom objects in real time!

Next Article – https://wp.me/p6xoZs-3G