coral image processing

To create model/graph

• Install docker
• Install tensorflow image: docker run -it gcr.io/tensorflow/tensorflow:latest-devel
• Exit docker by typing exit then create a folder in $Home called tf_files
• Create a folder called photos within tf_files folder
• Place all images in species labeled folder under the photos folder
• Open terminal and link images to docker instance: docker pull tensorflow/tensorflow
• Get the latest training code:
  • cd /tensorflow
  • git pull
• Run training:
  • python tensorflow/examples/image_retraining/retrain.py --bottleneck_dir=/tf_files/bottlenecks --model_dir=/tf_files/inception --output_graph=/tf_files/retrained_graph.pb --output_labels=/tf_files/retrained_labels.txt --image_dir /tf_files/photos
  • Grab the retrained_graph.pb and retrained_labels.txt files from $Home/tf_files/ and place into /Users/PROJECT

To classify

• Put all images to be classified in a folder called upload under /Users/PROJECT
• Open terminal
• Go to /Users/PROJECT
• Run the following commands:
  • source ./tfenv/bin/activate
  • python cnn_classify.py
• The program
• moves the images to a folder named after the coral species name
• Inserts two rows per image in the coral.db sqlite db: one row for highest scored species and another row for second highest score distinguished by a column called rank. Other db columns are as follows:
  • File: name of the file Species: name of the species
  • Date: date of the classification
  • Score: confidence level in percent decimal format
  • Rank: 1 for highest confidence, 2 for second highest confidence
• deactivate