object_detection

This is a collection of all of the openVino, roboflow, ultralytics tools. It is not complete and will be added to.

DEPENDENCY ISSUES:

- autodistill_all code has had issues newest version of ultralytics 
    - autodistill uses ultralytics version 8.0.81
    - yolov8 uses ultralytics version 8.0.238 
- Recommended to create seprate enviroments for autodistill and ultralytics 
- utils/scrape_yandeximages.py: may need to tweak elements depending on browser and location.

Directories:

autodistill_all:

• This directory contains scripts and utilities for running AutoDistill
• autodistill_utils:
  • display_utils.py: utility to display graphs in python
  • vid_to_img.py: convert video to img frames
• run_autodistill.py:
  • create distilled models with DINO or other LFMs
  • helps with autolabeling in theory

utils

• This directory contains general utility files that are not specific to any system
• scrape_googleimages.py: uses Selenium to scrape google images
• scrape_yandeximages.py: uses Selenium to scrape yandex images (yandex has provided the most relevant images results imo)
• bags_save_msg.m: MATLAB saves frames of selected bag topic.

yolov8

• This directory contains all files related to yolov8 specifically
• config.py
  • sets up paths and configurations
  • NOT REQUIRED MAY BE USEFUL IF YOU WANT A SINGLE PLACE TO DEFINE SUCH THINGS
• initial_setup.py
  • downloads coco trained yolov8 model
  • ensures all packages and requirements are functional
• model_benchmark.py
  • tests model performace
  • compares model stats for different model formats
  • run after training to decide which model format you want
• model_eval.py
  • evaluates single model on specified dataset
  • run after training to verify model performance matches training stats
  • ideally use new dataset
• model_quantize_nncf
  • creates quantized int8 model in openvino format (.bin, .xml)
  • significantly reduces model size and slightly reduces inference time
• model_setup.py
  • NOT USED ATM
• model_train.py
  • trains ultralytics yolov8 model on roboflow dataset
  • functions to download roboflow dataset and to train yolo model on roboflow dataset
• notebook_util.py
  • various functions to aid display and evalation
  • USED BY initial_setup.py