Object detection using YoloV5 (Part 1)

Annotation for object detection (YoloV5):

In this tutorial, we are going to see that how to prepare the data set for the object detection using YoloV5. And in the next tutorial, we will learn that how to train the YoloV5 model on the custom data set.

Annotation is a method for creating the bounding boxes around the objects in the images. Bounding boxes is one of the most popular and recognizable image annotation methods used in deep learning and machine learning. Using the image annotation method, we create the outlines (bounding boxes) around the objects in the images as per the requirement of our project. For example: if we want to detect the car. So we make a bounding box around the car like there is one image and there are 2 cars in the image. So we create the 2 bounding boxes around the car. 

Download the data set before annotation, we are using here a two wheeler and four wheeler vehicle data set. So you can download the car, bike and scooter dataset, then you need to annotate. I'll provide the dataset for testing.

Go to the app.roboflow.com website using the above link and create an account with your personal Gmail id or you can create an account using the GitHub account.

We can apply annotations using different tools like LabelImg. But personally, I suggest a website for annotating the data set called app.roboflow.com. This website helps you to annotate the data set in different formats like txt, csv, json etc. You can visit the website using the link below.

Annotation Link: https://app.roboflow.com/projects

Let's start :-

Create a new project. At time of creation, simply give your project name, project type will be Object Detection (Bounding Box) and Enter your Annotation group. Then simply click on the create project.

Click on the Select Folder then a window will pop up then choose a folder where your data set has been kept and simply upload.

After uploading all files, click on the first file (image). Your annotation window will be opened.

Click on the create tool (square box shape) and start annotating on the images. Simply make a box around the object (car, bike, etc.) and give the class name for each annotation. You can see the example below:

After making a bounding box on one image, simply click on next and do it for all images.

After annotating all images, simply come back with the back button. Your all annotating images will be showing like below:

Then click on the finish uploading button and choose split images between Train/Valid/Test.

Make Train data 80%, Valid data 20% and Test data 0%. And click on continue.

If you want to add more images (data set), click on the Annotate button which is shown in the left side of the window.

Click on the Add images to add more data. Then a will pop up, select all files which you want to add in your data. Then annotate those images which were uploaded.
After clicking on the continue, your files start to upload. And after uploading it, click on Generate new version.

Then you can preprocess the images like apply filters, resize images etc. And you also apply Augmentation like flip, rotate 90, blur, crop, shear etc. It will generate more images from existing images.

Then finally click on the generate button to generate the data.

After generate the dataset, then click on export button which is shown at the top right corner of the window.

Then select a YoloV5 PyTorch Format from the drop down menu and this format is in the form of a txt file.

Select a download zip to computer and simply click on the continue.

Your file starts to export and download in your local system as a zip file.

Then extract the zip file in the local system.

When you open the file, you’ll get file structure like below:

1. images

1. Train

2. Val

2. labels

1. Train

2. Val

In the image folder, you will get two folders train and val.  Train folder stores the training images or you can say that 80% of the training images. Val folder stores the testing images or you can say that 20% testing images.

In the labels folder, you will get two folders train and val.  Train folder stores the training labels. or you can say that 80% training labels. Val folder stores the testing labels or you can say that 20% testing labels.

Label files are the annotation files and it stores image annotations and annotations are in the form of decimal numbers. And label files are the text file with txt extension.

These data are used to train the YoloV5 object detection model with a custom data set.

Download dataset of Two-Wheeler (Bike, Scooter) and Four-Wheeler (Car) : Click here to download

You can see the below video for annotating the dataset of Two Wheeler (Bike, Scooter) and Four Wheeler (Car).

Video Tutorial

Thank You !!!!!!!