Detection of Turkish Coins using YOLOv5

The problem is to detect turkish coins and count the coin in each image accurately.

There are 5 Turkish coins labeled from 0 to 4. Here are the classes

• 0 is 5 kuruş
  
• 1 is 10 kuruş
  
• 2 is 25 kuruş
  
• 3 is 50 kuruş
  
• 4 is 1 lira
  

Data Collecting

First, we took pictures of different numbers of coins and from different angles to detect the coins.
After we used LabelImg and yolo format to label each image.
We added few examples of images that we took. All the images and annotations can be found under data folder.

Training the YOLOv5 Model Using PyTorch

Yolov5 is a higly efficient and accurate model among the other object detection models.

Algorithm	Inference Time(sec)	mAP
DETR-DC5	0.097	49.5
YOLOv5x6	0.045	55
Faster R-CNN	~0.2	48.1

Table 1. Comparison of Object Detection Models (YoloV5, Faster R-CNN, DETR)

To train the model with our dataset we used original documentation. Train Custom Data

Data Augmentation with Albumentations

Albumentations is a computer vision tool that boosts the performance of deep convolutional neural networks.

We used few data augmentation methods to enlarge our dataset.

• Random Brightness Contrast
• Gauss Noise
• ISO Noise

For more detail you can check the augmentaiton file and Bounding Box Augmentation documentation.

Contributors ✨

Thanks go to these people.

Made with contrib.rocks.