Fig. 7: SleepXViT training process.
From: Explainable vision transformer for automatic visual sleep staging on multimodal PSG signals
The SleepXViT training process consists of two stages: Intra-epoch ViT and Inter-epoch ViT. a The Intra-epoch ViT training, which employed the Vision Transformer (ViT) architecture34 to learn the embedding of the epoch image. b The inter-epoch ViT training, uses the frozen Intra-epoch ViT to generate embedded vectors. These vectors are then input into another transformer encoder, which generates the consecutive l epochs from referencing the adjacent epochs.
