Sum12 Multi-Scale Branching Encoder for Few-Shot Image Classification

The Sum12 architecture improves Prototypical Network embedding stages using seven parallel convolutional branches with kernel sizes summing to 12. This lightweight design captures diverse spatial structures within a unified multi-scale feature space for improved few-shot learning performance on benchmarks like CIFAR-FS and miniImageNet.