The idea here is to store previously trained models and provide the basic structure for training and testing environments.
Your local test/train directories can remain messy, but when you add your trained models and data here, make sure it remains organised.
-Contains full imagesets with labels
-(not split into val-train)
-Interesting results of previous test runs
-You can add images of detections both good and bad, graphs, etc.
-Contains the basic structure you need for testing/running a model
-testvideos folder contains videos that can be used to test the models
-If you want to test on an imageset, you can copy images from TRAINED_MODELS/[modelname]/data/ or IMAGESETS
-Copy the contents in a local directory and run testing there!
-yolo_detect.py is the script for running a test. For an example command, check runModelCommand.txt
-Contains the basic structure for training a model
-Copy the contents in a local directory and train there!
-Do not add any training data here
-The data_example folder here is just an example to show what the data structure should look like. You will replace it in your LOCAL train directory.
-train_val_split script splits your training data into train and validation folders
-trainCommand.txt has an example command to train a model
Your TRAIN folder should contain:
a "data" directory with "train" and "validation" inside
data.yaml file (set it up according to your data)
a model to fine-tune or if starting from scratch, the command should automatically download a model from ultralytics github
don't worry about cross_validate.py, you don't need it
-Store previously trained models here
-Within their respective directories, add the data, runs, .yaml and the actual .pt models