marcoscravero2175 / hair_style_generator

This is a Pix2Pix implementation that tries to generate a haircut based on a drawing

hair_style_generator

This is a Pix2Pix implementation that tries to generate a haircut based on a drawing

Purpose

Just have fun using a machine learning algorithm to solve a trivial problem

Setup

1. Import the notebook into Google Collab.
2. Create the following folders in your Google Drive:
   • /content/drive/My Drive/pix2pix/jmamoon/hairstyle_filtered/input
   • /content/drive/My Drive/pix2pix/jmamoon/hairstyle_filtered/output
   • /content/drive/My Drive/pix2pix/jmamoon/hairstyle_filtered/generated_training
   • /content/drive/My Drive/pix2pix/jmamoon/hairstyle_filtered/generated_testing
   • /content/drive/My Drive/pix2pix/jmamoon/hairstyle_filtered/checkpoints
3. Import the dataset from https://drive.google.com/drive/folders/13eFWEYs-onnpE__gZkTXz-IeEcJzwMJG
4. Run the notebook

About the dataset

I used the Figaro 1k dataset to generate the input and output images. You can see more information about Figaro 1k in http://projects.i-ctm.eu/en/project/figaro-1k

Problems I faced

Dataset had too much variety I needed to filter to improve generation, but it reduced my dataset amount, so, I think it reduced the data quality.

What I've learned

Dataset determines how good is your neuronal network going to work.

Conclusion

Results were not as good as I expected, but I think It can be improved by adding more images in the dataset and preparing the input images in a better way.