Add functionality to build yolov5 yaml

[dsikka]

Summary:

• Provide the functionality to build a yolov5 yaml file, given the user provides an input directory to the data arg.
• The following assumptions were made about the directory structure and a sample directory structure is shown below

Assumptions:

• The directory passed in contains an images subdirectory, a labels subdirectory, and a file called classes.txt which contains the classes, ordered by class id
• Each of the images and labels subdirectories contains training, testing and validation folders
• The labels should already be in the format expected by yolov5
• yolov5 assumes the train, test and validation directories have identical names in both the images and labels folders

Sample Directory Structure:

- data_for_training/
    - labels/
        - train/
        - val/
        - test/
    - images/
        - train/
        - val/
        - test/
    - classes.txt

Example Inputs/Outputs:

Input directory structure:

dsikka@quad-mle-2:~/some_dir/some_dataset$ ls
classes.txt  images

classes.txt:

dog
cat 
house 
park

images dir:

dsikka@quad-mle-2:~/some_dir/some_dataset/images$ ls
test  train  val

yaml file produced:

path: some_dataset/
train:
- images/train
test:
- images/test
val:
- images/val
names:
  0: dog
  1: cat
  2: house
  3: park

Testing:

• Locally tested training_aware runs by passing in a local directory to the data arg. yaml file produced is correct and training runs to completion

[rahul-tuli]

Loogs good pending comments!

[bfineran]

LGTM pending rahul's comments

[dsikka]

Updated and retested.