Running pre-trained TF Lite models for object detection. You either have to install Tehsorflow or Tensorflow Lite (tflite_runtime) and OpenCV (opencv-python). These scripts also run a lot faster on a ARM device, for example, a Raspberry Pi 3B or 4B.
There are three models available here (downloaded from Google):
- SSD-MobileNet V1
- EfficientDet-Lite0
- YOLO V5
All three are trained with the COCO dataset (labelmap.txt is the label list). This is mainly a demostration of how to get the possible things as well as their location from the model.
TF_Lite_Object_Detection.py can use either SSD or EfficientNet to process a still image, and TF_Lite_Object_Detection_Yolo.py is the YOLO version. TF_Lite_Object_Detection_Live.py use live USB cam images with SSD or EfficientNet (press q).