Skin-detection(Using GCN)

• SFA (Skin-nonSkin dataset) - Citation : CASATI, J. P. B. ; MORAES, D. R. ; RODRIGUES, E. L. L. . SFA: A Human Skin Image Database based on FERET and AR Facial Images. In: IX Workshop de Visão Computacional, 2013, Rio de Janeiro. Anais do VIII Workshop de Visão Computacional, 2013.
• If you want to use this Dataset, go to URL : http://www.sel.eesc.usp.br/sfa/ (I don`t upload the SFA file Because it could be a problem)
• I am currently writting a paper with this code, If it is completed, I will also post my thesis
• If you use it commercially, please contact my email, If you use it in github or other open source in a project, please provide a source.

• [email protected] & [email protected]

• openCV
• numpy
• matplotlib
• pandas
• tqdm
• pytorch
• hashlib
• json
• os
• torchvision
• argparse
• seaborn

• Step 1. Make_dataset (https://github.com/YangChangHee/SKIN_GCN/blob/main/Make_dataset/step%201.%20make%20dataset.ipynb)

Check SFA file and make adjacency Matrix

3-dimension RGB =>(Convert) 5-dimension R,G,B,Cb,Cr

adjacency Matrix =>

• Step 2. make_parameta (https://github.com/YangChangHee/SKIN_GCN/blob/main/make_parameta/step%202.%20make_parameta.ipynb)

I put a neighboring matrix in class

Change Markdown to code to see how dimensions decrease

If you look at In[5], In[6], (making random matrix)Apply parameta

• Step 3. test_basic model (https://github.com/YangChangHee/SKIN_GCN/blob/main/make_parameta/step%203.%20test_basic%20model.ipynb)

The GPU used is the RTX-2080TI

I Checked Whether Step1 and Step2 can be used properly.

• Step 4. Large data make datafile (https://github.com/YangChangHee/SKIN_GCN/blob/main/Make_dataset/step%204.%20large%20data%20make%20datafile.ipynb

The data we will use should be refined. So we use a new data form , .npy

The back is a method of applying the loaded data to the model

• Step 5. load data and model define (https://github.com/YangChangHee/SKIN_GCN/blob/main/Define_Model/step%205.%20load%20data%20and%20model%20define.ipynb)

I tried increasing the model hidden dims

Tqdm was used to determine how long the traing took

Finally, we applied it to the general image.

• Step 6. GCN deep (https://github.com/YangChangHee/SKIN_GCN/blob/main/Define_Model/step%206.%20GCN%20deep.ipynb)

Increasing model hidden dims(2 => 3)

• Step 7. GCN Result (https://github.com/YangChangHee/SKIN_GCN/blob/main/%EB%B3%B4%EA%B3%A0%EC%84%9C/Step%207.%20GCN%20Result.ipynb)

Hidden dim 1 => [10], Hidden dim 2 => [15, 20, 25, 30], Hidden dim 3 => [20, 30, 40, 50]

Result

• Step 8. GCN Result (https://github.com/YangChangHee/SKIN_GCN/blob/main/%EB%B3%B4%EA%B3%A0%EC%84%9C/Step%208.%20GCN%20Result.ipynb)

Add Visualize Function

Result (Change Imagesize [imagesize] => [180,180])