In this repository, we are going o learn how to train model on yolov4mini for your own objects from video file. Following codes mainly comes from github.com/moises-dias/yolo-opencv-detector with mainly changes in creating dataset, usage of video and FPS increase changes for realtime detection. Thanks to the moises-dias. You can find video of his work in following link.
https://youtu.be/RSXgyDf2ALo?si=uhzuU-ZM-UmqqzEM
I recommend using environment for libraries. Go to https://www.anaconda.com/ and downloand Anaconda Navigator After setup, run Anaconda Navigator
On Anaconda Navigator, go to Environments tab.
Create a new environment on Python 3.11.* and name it.
After creating environment, run Anaconda Prompt
If you see (base) in the start of the command line, you must open the environment first.
You can use conda env list command to see environments on your computer. You must see your newly created environment in this list.
Use following command to open your newly created environment. Make sure to chang last word with your environment name.
conda activate YOUR_ENVIRONMENT_NAME
First, clone repository.
Then go to file using cd object_detection_training_yolov4mini in your terminal. (Do not shut down the terminal)
Use python install -r requirements.txt code in terminal to download necessary libraries.
Make sure that your video is inside the 'videos' folder with name 'video.mp4'
Run generate_dataset_from_video
Examine images folder when code running. If you decide you have saved enough files, press 'q' to finish the process (generally 300 to 1000 images depend on the classified objects)
After creating data, go back to your terminal and run jupyter notebook . command.
In jupyter notebook, run label_dataset.ipynb according to instructions that given in the file.
Trainign can be take 5 hours depends on your dataset and labels. But within 15 to 30 minutes, you can use newly created file on drive/yolov4-tiny yolov4-tiny-custom_last.weights with a certain degree of coherenc to object detection.
Copy yolov4-tiny-custom_last.weights to your folder which is object_detection_training_yolov4mini
run yolo_opencv_detect.py