Finger Detection using Pytorch

We use the Convolutional Neural Network model to identify the number of fingers ✋ shown from the photograph.

Refrence:Pytorch

Requirements

 1. Pytorch
 2. torchsummary
 3. Pygame
 4. Opencv

Usage

Prepare Training Dataset

Create folder named data . Then create folders according to your class-label names(n folders for n classes).Then place all images into the corresponding folder(class-label).

To capture images run python data_collection.py specify the correct folder to the classes accordingly. Then, run mask.py to achieve binary mask that is defined by grayscaling and blurring the image.

Folder structure

Root
│
├── data
│    ├── train
│    │     ├── ONE
│    │     │    ├── ONE_0.png 
│    │	   │	├── ONE_1.png
│    │	   │	└── ...
│    │	   ├── TWO
│    │     │    │    
│    │	   │	├── TWO_0.png
│    │	   │	└── TWO_1.png
│    │	   ├──	.
│    │	   └── 	.
│    │				
│    │     
│    └── test  (model)
│          ├── ONE
│          │    ├── ONE_0.png 
│    	   │	├── ONE_1.png
│    	   │	└── ...
│    	   ├── TWO
│          │    ├── TWO_0.png 
│    	   │	└── ...
│    	   │	
│    	   └──	FIVE
│    	    	│
│               ├── FIVE_0.png
│               └── ...
│
│
├── Finger Recognition.ipynb
└── model_chk.pt 

Larger the number and variation in images, larger the accuracy of classification.

Convolution Neural Net

----------------------------------------------------------------
        Layer (type)               Output Shape         Param #
================================================================
            Conv2d-1           [-1, 32, 66, 66]             896
              ReLU-2           [-1, 32, 66, 66]               0
         MaxPool2d-3           [-1, 32, 33, 33]               0
            Conv2d-4           [-1, 64, 35, 35]           18496
              ReLU-5           [-1, 64, 35, 35]               0
         MaxPool2d-6           [-1, 64, 17, 17]               0
            Conv2d-7          [-1, 128, 19, 19]           73856
              ReLU-8          [-1, 128, 19, 19]               0
         MaxPool2d-9            [-1, 128, 9, 9]               0
           Linear-10                    [-1, 6]           62214
================================================================
Total params: 155462
Trainable params: 155462
Non-trainable params: 0
----------------------------------------------------------------

Training

Run the Finger Recognition.ipynb jupyter notebook to start the Training.

Once the training is completed the result is stored the main folder as model_chk.pt

Testing and Performance:

Average accuracy on the test set is 97.28448438289819

Loss

Output

Total CNN Architecture graph

Conv1 visualization

ToDo

• Train.
• Test.
• Predict.
• Save/Load checkpoint.
• UI for input, output and to download checkpoint.