It can be seen from the previous screenshot that even in the case of varying object sizes and also objects with small sizes, the two-stage model of Faster R-CNN predicts accurately. Now, we will show how to run a similar prediction using TensorFlow. Let's begin by cloning a repository, as it will contain most of the required codes:
git clone https://github.com/tensorflow/models.gitcd models/research
After we have cloned, we will set up the environment. We will first download a pre-trained model from TensorFlow model-zoo:
- For macOS X:
curl -O http://download.tensorflow.org/models/object_detection/faster_rcnn_resnet101_coco_2017_11_08.tar.gztar -xvf faster_rcnn_resnet101_coco_2017_11_08.tar.gz
- For Linux: ...
