Training_model_notes

# Info from Kevin meeting 22nd of Mar 2022

# New weights from Enhance and not enhance

+ model_final_non_enhanced.pth :

    location : googledrive BGNN/BGNN_workflow/Model_weights/Drexel/
    https://drive.google.com/file/d/1YHGUkY1EUAfeHHd3crZToQMIrseW6INo/view?usp=sharing
    new weights form kevin work without the CLAHE (histogram equalization enhancement)
    
+ model_final_enhanced.pth :

    location : googledrive BGNN/BGNN_workflow/Model_weights/Drexel/
    https://drive.google.com/file/d/13pa5E5odN_gWNZYkA12u8ZEnEjzWGxFL/view?usp=sharing
    
    new weights form kevin work with the CLAHE (histogram equalization enhancement)
    
    
    
# Training

Dataset coco jason *.json
To generate the "cococ.json" dataset format use makesense.ai
1- load the image
2- load the txt file with cataegory/class/instance (i.e.: fish, eye, ruler, two, three) or create manual in the API
3- label your images
4- export the coco.json file (contained the annotation of your images)

Use train_model.sh
change or create a file like "training_data_all.json"
change the prefix file overall_prefix.txt
checkpoint resume true