Skip to content

Latest commit

 

History

History
98 lines (64 loc) · 2.43 KB

README.md

File metadata and controls

98 lines (64 loc) · 2.43 KB

Hand_Sign_Music_control

Most of the time while working we dont have space on screen for app to control the background activities like music. This project is an example of use of handsigns as input for web-automation.


Requirements for usage

  • Python 3.8 or above

    Python libraries:

    • Pytorch
    • CV2

      pip install opencv-python

    • Selenium

      pip install selenium

  • Geckodriver (present in the repository)

  • Webcam

Usage


Requirements for training CNN model

  • Python 3.8 or above

    Python libraries:

    • Pytorch
    • CV2

      pip install opencv-python

    • Matplotlib

      pip install matplotlib

Training Stats


Usage

Run the code using command line

use of plane background is recommended

A webcam window will open.

Instructions Image

Make these handsigns in the visible region of output of webcam.

Until the open browser sign is made all instructions will be ignored.

Demo video


Training Stats

Input image size=28x28x1

DataSet:

Train Set = 9000 Images
Dev Set = 1800 Images

Architecture used:

(FC-Fully Connected Dense Layer,F-Features,S-Strides,C-Channels)
Conv Layer - F(3x3) S(1x1) C(1,64)
BatchNorm 
Max Pool - F(2x2) S(2x2)
Conv Layer - F(3x3) S(1x1) C(64,128)
BatchNorm
Max Pool - F(2x2) S(2x2)
FC(5*5*128,128)
BatchNorm
FC(128,9)

Optimizer Used: ADAM

Learning Rate(alpha)=0.01
Decay rate=(0.95^i)    {i is epoch number}

Number of Epochs = 10

Mini Batch Size = 1000

Cost stats--time taken in seconds

Cost Graph

Train Set Accuracy = 100%

Dev Set Accuracy = 99.88%


Contributors

Atharva Kathale