This repository contains a python script and a few Object Detection models utilizing TensorFLow Lite. These models are placed in two folders i.e. 'custom' and 'pretrained'. The models located in the 'custom' folder are created using the Tensorflow Lite Model maker and can be trained to detect various objects from a desired dataset. In this case, that dataset happens to be field elements from the 2023 FRC Competition Charged Up.
The models in 'pretrained' folder are downloaded from coral.ai website. These pretrained models are trained with COCO dataset to detect 90 types of objects.
The python script can be used to run a custom as well as a pretrained model. It also supports the use of a Google Coral USB accelerator to speed up the inferencing process or any other Edge TPU containing device.
The training is done through a Colab notebook which is an interactive Python notebook accessible through a web browser. It makes use of Tensorflow Lite Model Maker to create custom models through Transfer Learning.
The link to the notebook is here
The annotated data set created for this project is here. Annotations were performed with the help of roboFlow.
The notebook provides a framework to create and download a custom model for object detection using any custom dataset of choice. From the notebook, the corresponding models and label files can be downloaded and uploaded into this project.
The packages and libraries required to run this file can be installed through bash script by running the command 'sudo sh setup.sh' in terminal. Alternitively there is a 'requirements.txt' that can be utilized for installation, or the direct dependencies may be manually installed.
Run the python file using the command 'python3 detect.py'
You can use a Pi camera or a USB camera with your Raspberry Pi to run the python file 'detect.py'. The python script also supports Google Coral USB Accelerator. If you want to use Coral Accelerator and Edge TPU framework, ensure the appropriate procedure is followed in the script labeled detect.py.