Setup

Make sure to have Python 3.9 and Node.js 18.0.0 installed

$ python3.9 -m venv venv
$ . venv/bin/activate
$ pip install -r requirements.txt
$ cd cnn_code
$ pip install -r requirements.txt
$ wget https://data.vision.ee.ethz.ch/cvl/rrothe/imdb-wiki/static/gender.caffemodel
$ cd ..

Then, in one terminal:

$ cd server
$ flask run

…and in another:

$ cd frontend
$ npm install
$ npm start

Web UI

1. In one terminal, run flask run in the server/ folder
2. In another terminal, run npm install and then npm start in the frontend/ folder

How to run tutorial at the moment

pip install -r requirements.txt Download the gender.caffemodel in the link below and place in cnn_code directory

Resources

• https://pysource.com/2019/05/28/face-swapping-explained-in-8-steps-opencv-with-python/
• https://pypi.org/project/dlib/
• https://pytorch.org/tutorials/beginner/finetuning_torchvision_models_tutorial.html

Datasets

• https://www.kaggle.com/datasets/dadajonjurakuziev/movieposter
• https://www.kaggle.com/datasets/xuejunz/film-posters-information
• https://www.kaggle.com/datasets/spiyer/old-film-restoration-dataset
• https://www.kaggle.com/datasets/cashutosh/gender-classification-dataset/code (could be used for training a model for sex classification)
• https://www.kaggle.com/gpiosenka/gender-classification-from-an-image (smaller sex data set, in repo currently)
• https://drive.google.com/drive/folders/1-0YhtXe_oE2ei0R471X33a_NJyY5dVge

Meetings w/ Ta

April 15

• look into pretrained facial feature models
  • https://pypi.org/project/dlib/
• guessing sex
  • might need to train a secondary model
  • start by not worrying about this, mismatches are ok
  • instead get it working good first
  • later add a second level of detection
  • also don’t worry about glasses too much
• focus on blending well with skin instead (/ “does it look like it’s the actual person’s face”)
• worth a try with skin tone
• TA will look into preexisting models
• facial expression / emotion may impact results
  • consider finetuning pre-existing models
    • take an existing model and then add a few layers onto the end of it, highly specialized to their dataset
    • train the new layers using a small dataset
    • start w/ pytorch tutorials (which go step by step through this process)
    • https://pytorch.org/tutorials/beginner/finetuning_torchvision_models_tutorial.html
• do not worry if it doesn’t work, we’re supposed to explore / do something cool
  • more important: did we put time/effort in?
  • did we try different things?
  • writeup is very important