YoloV5 Training Module

This module schedules a training of YoloV5 with the corresponding training and testing BQ datasets. The labels from each dataset come from GObjects.

Parameters

• Training Dataset: the dataset to be used for training
• Testing Dataset: the dataset which will be used to produce validation metrics
• Classes .txt: A .txt file which maps each GObject annotation to a class (integer) which ranges from 0 to n-1 where n is the number of classes. An example will be given below
• Starting Weights: (OPTIONAL) A pytorch file to begin training from. If this is empty, Yolo will be trained from scratch
• Training Set GObjects: the name of the highest-hierarchical GObject. If an image in the training set does not have a Gobject with this name, the image will not be used in training.
• Testing Set Gobject: the name of the highest-hierarchical GObject. If an image in the training set does not have a Gobject with this name, the image will not be used in testing.
• ClearML Access Key: (OPTIONAL) A ClearML access key to track the experiment
• ClearML Secret Key: (OPTIONAL) A ClearML access key to track the experiment
• Epochs: How many epochs to run the training loop

Example Classes .txt

If we have a dataset with 3 classes which are "person", "car" and "dog", the .txt file will look like this:

 0 person
 1 car
 2 dog