elhamod / phylonnbutterflies Goto Github PK

This repo is for Fish classification.

To train:

• Have your data with metadata.csv in a directory. A sample file can be found in HGNN/analyse/HGNN_repo_artifacts.zip. Your image data needs to be in the torchvision ImageFolder format, with train/val/test subfolders.
• Create a config file using HGNN/train/ConfigParserWriter-SelectExperiments.ipynb
• train a model using HGNN/train/train.ipynb.

To view results:

• analyze the data using jupyter notebooks HGNN/analyse/Analyze experiments.ipynb.
• Once trained and analyzed, an experiment folder with name / will have been created, with a "models" folder that has all the trained model, a "results" folder with the analysis.
• You can also look at HGNN/analyse/Analyze multiple experiments-*.ipynb for seeing the results in our MEE paper
• To see Saliency map results, you can look at HGNN/analyse/Saliency Maps - *.ipynb files.