Text this: Adaptive multi-scale feature extraction and fusion network with deep supervision for retinal vessel segmentation.