Sample TFRecords and pbtxt file

TensorFlow Object Detection API requires the data to be packaged in TFRecords at least if you are using the API in a non-eager mode. It also needs a .pbtxt file that defines the integer mapping of the class labels.

TFLite models for pet detection

The original model is fine-tuned from an SSD MobileNetV2 model. The training notebook is available here. The attached TFLite model was quantized using dynamic-range quantization.