Silas Gifford Honors Thesis

Here is the code, images, and data on the populations of interest in Point Reyes National Seashore.

Basic Info

1. Harbor Seals
   • Date Range:
     • Pupping Season: March to May
       • Age classes: "adult" and "pup"
     • Molting Season: June and July
       • Age class: "adult"
   • Drakes Estero
   • Double Point
   • Tomales Bay
   • Tomales Point
   • Point Reyes
2. Elephant Seals
   • Date Range:
     • Pupping Season: November to March with peak between 1/26 and 2/2
       • Age classes: "bull" (adult male), "cow" (adult female), and "pup" (baby)
   • Point Reyes Headlands
   • Drakes Beach
   • South Beach

Recreate my work by...

• Using the full images pulled from Google Earth found under Images
• Tile these images with Code/Tile_Harbor_Seals.ipynb and Code/Tile_Elephant_Seals.ipynb
• Create your own annotations via https://www.robots.ox.ac.uk/~vgg/software/via/app/via_image_annotator.html or use mine found in Data/Harbor_Seal_Anno_csv.csv (a single csv file) and Data/Elephant_Seal_Anno_csv (a folder of csv files).
• If you're using your own annotations, use this repo https://github.com/martinzlocha/anchor-optimization/ to get anchor configurations, or use mine at Data/habor_config.ini and Data/elephant_config.ini
• Convert annotations into the right format for Retinanet using Code/Harbor_Seal_Annotations_Right_Format.ipynb and Code/Elephant_Seal_Annotations_Right_Format.ipynb
  • Please note that this code will give you randomized training, validation, and testing sets. If you want to train/test on specific imagery, you must change the code.
• Make sure you have Retinanet installed: https://github.com/fizyr/keras-retinanet
• To train on all harbor seal tiles, use Code/Harbor_Seals_Run_Model_Full.ipynb
• To train on harbor seal beach-only tiles, use Code/Harbor_Seals_Run_Model_Beach.ipynb
• To train on elephant seal beach-only tiles, use Code/Elephant_Seals_Run_Model.ipynb
• Use Code/harbor_seal_visual_check.ipynb and Code/harbor_seal_visual_check.ipynb to visualize model output or view model output in Data/model_output