In this exercise, we will use the TensorFlow object detection API to train a custom object detector using four different models. Google Colab is a VM that runs on the Google server, so all of the packages for TensorFlow are maintained and updated properly:
|
# |
Model |
Feature Extractor |
|
1 |
Faster R-CNN |
Inception |
|
2 |
SSD |
MobileNet |
|
3 |
SSD |
Inception |
|
4 |
R-FCN |
ResNet-101 |
Note that at the time of writing this book, the TensorFlow object detection API has not been migrated to TensorFlow 2.x, so run this example on the Google Colab default version, which is TensorFlow 1.x. You can install TensorFlow 2.x in Google Colab by typing %tensorflow_version 2.x—but then, the object detection API will result in an error. The demo exercise has TenorFlow version 1.14 and numpy version 1.16...
