Yolov5-Course-Design

Author:Ziyan Jiang([email protected]), Ying Zhu, Jiayi Jin

Yolov5(You Only Look Once Version 5) is an advanced object detection model, . YOLOv5 builds on the YOLO (You Only Look Once) series of models and is capable of performing object detection tasks in a single forward pass, simultaneously predicting multiple objects in an image along with their bounding boxes and classes.

Here are the steps to run the task.

Resize the Origin Images

First, we resize the images that we take to 640*640, to make it sutiable for the following task, you could see the code in script resize_images.py. You could use it by just changing the data location.

Use LabelImg to label the data

To change the origin images to annotations, we use the tool LabelImg to mark the data, the code is in the folder labelimg.

Use Yolov5 to train the model

We use yolov5 to train the model, the image in folder yolo/course_design/images, the labels in folder yolo/course_design/labels, you can change training set, validation set, number of classes, class names in info.yaml.

To train the model, adjust the config in run_course_design.sh, and input following code in terminal:

sh run_course_design.sh

After training, you will get the result in folder yolo/runs/train.

Visualizing results with PyQt5

We use PyQt5 to visualize the detecting results. You can read the detailed code in the folder yolo/init_ui.py.

We realize the following functions:

1. Input image: by clicking the "Input Image" button, you can choose a image from your computer, and at the same time show the image in the right layout.
2. Run detection: by clicking the green button on the top of the left layout, you can start detecting the image you input.
3. Show result: by clicking the "Show Result" button, you can see the detection image as well as all the categories in the right layout.
4. Clear image: by clicking the "Show Result" button, you can clear the result.
5. Show minimized: by clicking the yellow button on the top of the left layout, you can minimize the window.
6. Close: by clicking the yellow button on the top of the left layout, you can close the window.