Code and guidance (written by Sicily Fiennes, with the assistance of Sam Watts) to identify traded animals in highly occluded contexts, based on images.
- Using the merge function in Python to set up a match-mismatch survey
- Testing for differences between match/mismatch questions and plotting phylogenies using ggfree
- Glossary of machine learning terms
- Hardware requirements: setting up Python and downloading TensorFlow
- Running code on University High Performance Computers
- Object Detection using the MegaDetector to localise and extract bird crops
- Data pre-processing: image augmentation as a method of class balancing
- Training convolutional networks: training the models for species identification
- Ensembling models
- Evaluating model performance using cross validation
- Building a binary model
- Superimposing uncaged images with caged masks in the foreground
Our work flow for the classification of 37 bird species
- Machine Learning Mastery - Dr. Jason Brownlee
- Deep Lizard
The website for this work can be found @ https://sicily-f.github.io/cagedbirdID/, which has more rationale for the project. For more information about our methods, processes of deduction and tool selection please contact [email protected]. If you have a question related to the material presented here, please create a New Issue under the ‘Issues’ tab above. If you can specify the name of the notebook which your question is related to, that would also be great.