Ageing Synthesis by CVAE / CVAE-GAN

Ageing synthesis on fundus or face images via Conditional VAE (CVAE) or CVAE-GAN model. In CVAE-GAN architecture, the training process is two-stage. You can use a pre-trained classifier or regressor to guide the generator, or you can also train the provided model.

Data preparation

The training, validation and testing data should be saved as json format under the ./dataset file.

dataset

train.json

validation.json

test.json

A json file should contain a list of dictionaries, each of which includes at least two pieces of information: "img_path" and "target" . A example is as following:

[
    {
        "img_path": "./.../1.png",
        "target": 45,
    },
    {
        "img_path": "./...2.png",
        "target": 52,
    }
]

where "target" is the training label (age in this task).

Run

cvae.py: execute the generation task, train and test, via Conditional VAE architecture.
main.py: execute the generation task, train and test, via CVAE-GAN architecture.
classification.py: train and test a classifier.
regression.py: train and test a regressor.

Oculins/CVAE-GAN

Ageing Synthesis by CVAE / CVAE-GAN

Data preparation

Run