model-weights-disentangling-retinal-images

Model weights for the generative models trained for the paper "Disentangling representations of retinal images with generative models" arxiv.org/abs/2402.19186.

Naming convention for model weights: 

• gan_w_inversion_losses*: generative model only trained with gan inversion losses and no subspace losses
• age_cam_id_wo_dcor*: generative model trained with gan inversion losses and subspace classifier losses for patient age and camera encoding (the patient identity subspace has no constraints)
• age_cam_id_w_dcor*: generative model trained with gan inversion losses, subspace classifier losses for patient age and camera encoding and a distance correlation loss for subspace independence
• ethn_cam_id_wo_dcor*: generative model trained with gan inversion losses and subspace classifier losses for patient ethnicity and camera encoding (the patient identity subspace has no constraints)
• ethn_cam_id_w_dcor*: generative model trained with gan inversion losses, subspace classifier losses for patient ethnicity and camera encoding and a distance correlation loss for subspace independence