Due to the large file sizes of the model, the pre-trained models are not available in this GitHub repository but are shared via Google Drive instead. Below are instructions for using these pre-trained models to make predictions on .jpg files.
Visit the Google Drive link to the PyTorch model state-dicts directory via your web browser at https://drive.google.com/drive/folders/1Q9vr5Q0BueHD0Kal2pEmA-NWIbYZnJAL and click on the "Download All" button on the upper right corner (the downloaded files will be ~1GB in total).
After downloading and unzipping the files, cd into the "single-image-prediction directory". Inside, you will be able to find a subfolder with a script that can be used to run a model pre-trained on one of the different datasets evaluated in this paper:
Next, simply execute the script with an example image. For example, you can execute the following in order to execute the CORAL ResNet-34 trained on AFAD:
python afad-coral.py -i 35_years__1194-0.jpg
Class probabilities: tensor([[1.0000e+00, 1.0000e+00, 1.0000e+00, 1.0000e+00, 1.0000e+00, 1.0000e+00,
1.0000e+00, 1.0000e+00, 1.0000e+00, 1.0000e+00, 1.0000e+00, 9.9998e-01,
9.9991e-01, 9.9966e-01, 9.9883e-01, 9.9557e-01, 9.8391e-01, 9.4708e-01,
8.4080e-01, 5.8262e-01, 2.3082e-01, 5.3700e-02, 8.9853e-03, 9.9941e-04,
2.7141e-05]])
Predicted class label: 20
Predicted age: 35Note the class labels in the training sets start at 0, which is why the true age (Predicted age) is larger than the predicted label (Predicted class label).
If you get an error such as
FileNotFoundError: [Errno 2] No such file or directory: '../../training-results/afad-coral_seed0_imp0/model.pt'
please make sure that you cded into the directory where the Python script is located. Due to the relative link in that file (linking to the pre-trained models in the training-results directory), something like
python afad/afad-coral.py -i afad/35_years__1194-0.jpg
will not work (should be python afad-coral.py -i 35_years__1194-0.jpg).