ap-cricket

This repository is repsonsible for splitting a group of videos into images and then categorizing those images into blurry or not blurry by calculating the variation of the Laplacian. It also trains an object detection model using the PyTorch framework Detecto.

Getting Started

This project is built with Python v3.8.3 and uses the following dependancies. Please install the pacakages below before attempting to run any of the code in this respository.

OpenCV
pip install opencv-python
Numpy
pip install numpy
Matplotlib
pip install matplotlib
Progress
pip install progress
Pytest-Mocha
pip install pytest-mocha
PyTorch
torch==1.4.0 torchvision==0.5.0
Detecto
pip install detecto

Extracting frames from video

Place videos in the videos folder
Run extract_all_frames.py
Run extract_blurred_frames.py
Results will be in the images folder

Training object detection model

Label images using labelImg (Or any equivalent serivce that gives xml labels in PASCAL VOC format)
Put images for training dataset into training/images
Put labels for training dataset into training/labels
Put images for validation dataset into validation/images
Put labels for validation dataset into validation/labels
Run train_obj_model.py
Model and associated data will be stored in model/ and zipped at the root.

Using the model

Put testing images into testing/
Change image filename so that it corresponds to image name used in obj_detection.py
Run obj_detection.py

Testing Command

From the tests directory pytest --mocha
This will run all tests in the tests folder. Follow the naming convention when creating tests i.e. test_<your_name_here>. This allows the test runner to automatically detect test files and run them. The output should look like below:

=========================================== test session starts ===========================================
platform win32 -- Python 3.8.3, pytest-6.0.1, py-1.9.0, pluggy-0.13.1
rootdir: C:\Users\<YOUR_USERNAME>\Documents\ap-cricket\tests
plugins: mocha-0.4.0
collected 5 items

Package Check :: test_basic.py
    Python version
        ✓ Should have Python version >= 3.8
    Open CV
        ✓ Should have OpenCV version >= 4.2
    Numpy
        ✓ Should have Numpy version >= 1.1
    Progress
        ✓ Should have Progress version >= 1.5
    Matplotlib
        ✓ Should have Matplotlib version >= 3.0
                                         [100%]
============================================ 4 passed in 0.42s ============================================

Next Steps

Get metrics like the IoU, Precision and Recall
Find way to plot ground truth bounding box together with predicted grounding box
Autonomously extract the area inside the predicted grounding box and send to PyTorch 3D for extraction

Issues

!_src.empty() in function 'cv::cvtColor'

Resources

PyTorch models

simeon9696/ap-cricket