Recognizing Characters in Art History Using Deep Learning

This is the official github repository to the paper : https://dl.acm.org/citation.cfm?id=3357242

This repository is for the gender/sex detection/classification on faces/persons within art images (paintings, sculptures, art-works). Important things to note :

Installation and getting started:

• Run pip install -r requirements.txt.
• If you are using a GPU, check/edit the requirements.txt file to install tensorflow-gpu instead of tensorflow

1. Install cvlib : pip install --upgrade cvlib
2. Before running train and test, make sure you have downloaded and placed the following files as follows:
   a. cfg/yolov3.cfg
   b. model-weights/yolov3.weights
3. The data directories should be structured as :

data
├── train
│   ├── class0
│   ├── class1
├── test
│   ├── class0
│   ├── class1

4. There is one training script to generate all the models : train.py.
   • To generate model A and B:

   python train.py -d <path_to_dataset>

   • To generate model C:

   python train.py -d <path_to_dataset> -f True -mp <path_to_styled_model (model B)>

5. After the training, check if the appropriate models are saved in the respective folders (self-explanatory from the code)
6. Testing the model on random folder of images. Run

python test.py --testdir <path_to_testdir> --preddir <path_to_save_predictions> --model <path_to_trained_model>

7. The cams folder contains keras_cam.py. You can run it by using :

python keras_cam.py -m <path_to_trained_model> -t <path_to_testdir> -s <path_to_savedir>

Acknowledgements

This repo is adapted from the github repo : https://github.com/arunponnusamy/gender-detection-keras. The authors would like to thank Arun Ponnusamy for his amazing work and sharing the code to build and continue working together without "rediscovering the wheel".