Piano_Keyboard_Detection

Description

The piano keyboard detection is an object detection task that detects keyboard on a piano in pictures or videos. This repository includes a demo for building the keyboard detector using YOLOv5 model and SSD-MobileNet model.

Models

• YOLOv5 model
  
  • Go this project
    
• SSD model
  
  • Go this project
    → It is important to structure your directories strictly by following the guidelines well for this model.

Data

Dataset

We collected 2,605 images from divided in 5 technique related videos and we trained these data.
The class is defined as follow:

• keyboard

Data Collection

We collected data in a variety of ways.

1. We collected 2,605 images from related videos divided in five classes using OpenPose.(fps: 30)
   

2. We collected about 10 images by searching "playing piano" in google.
   

Data Labeling

1. YOLOv5

We used DarkLabel(Data Labeling Program operating in a Windows environment)

• You can download DarkLabel in Github (https://github.com/darkpgmr/DarkLabel)
  
• We labeled data following this file. => DarkLabel.pdf
  

2. SSD-MobileNet

We converted labeled data for YOLOv5 to Pascal VOC data. Then we converted Pascal VOC data to csv files to make tfrecord files.