YOLO Object Detection Algorithm Training and Evaluating Pipeline

Requirement:

1. System: Window. Linux and mac not supported (This is because the kernel and libraries in requirements.txt are specialized to window)
2. Dataset:

• Card dataset: https://www.kaggle.com/luantm/playing-card
• COCO dataset: https://cocodataset.org/#download
• Other dataset can be found here: *

3. VS Code (in order to run the jupyter notebook with python file: VS Code format)

File structure:

• yolo: Contains the core library of the algorithm.
  • __init__.py
  • const.py: Storing hyper params.
  • loss.py: Collection of loss functions
  • model.py: Collection of YOLO model and base conv model.
  • utils.py: Utillities methods.
• main_card_ds.py: This is the jupyter notebook (vscode format) to train the YOLO v1 algorithm on the solitare card detection dataset.
• main_coco_ds.py: This is the jupyter notebook (vscode format) to train the YOLO v1 algorithm on the coco object detection dataset.
• main_coco_ds.ipynb: This is the jupyter notebook to train the YOLO v1 algorithm on the coco object detection dataset on Google Colab (Training elsewhere will not work as expected).
• weights: contains the weights file
  • checkpoint9: The yolo v3 model with customization. Train for 2 epochs and 20000 step each epoch, with batch size 4.
  • checkpoint10: The yolo v3 model with customization. Train for 4 epochs and 20000 step each epoch, with batch size 4.

Run the training process:

• First, install the required libraries:

conda create -n tf
conda activate tf
pip install -r requirements.txt

• Next, go into one python file main_card_ds.py or main_coco_ds.py and experiment with the VS Code notebook.

• Alternatively, you can go to this colab notebook to see the training pipeline: https://colab.research.google.com/drive/1YamrFLBJ1K7Dmy7L3f7hLp0LeNaBFL4Y?usp=sharing. Remember to change the runtime type to GPU if you want to run the whole pipeline

Example:

• Here is the intermediary outputs in the convolution layers:

Reference:

@INPROCEEDINGS{7780460,
author={Redmon, Joseph and Divvala, Santosh and Girshick, Ross and Farhadi, Ali},
booktitle={2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)}, 
title={You Only Look Once: Unified, Real-Time Object Detection}, 
year={2016},
volume={},
number={},
pages={779-788},
doi={10.1109/CVPR.2016.91}}

• https://jonathan-hui.medium.com/real-time-object-detection-with-yolo-yolov2-28b1b93e2088