High-performance contrastive learning and ensemble swin transformers for scalable pose-invariant facial expression recognition | Synapse