Stitchable Neural Networks

The public model zoo containing enormous powerful pretrained model families (e.g., DeiT/Swin) has reached an unprecedented scope than ever, which significantly contributes to the success of deep learning. As each model family consists of pretrained models with diverse scales (e.g., DeiT-Ti/S/B), it naturally arises a fundamental question of how to effectively assemble these readily available models in a family for dynamic accuracy-efficiency trade-offs at runtime. In this work, we present Stitchable Neural Networks (SN-Net), a novel scalable and efficient framework for model deployment which cheaply produces numerous networks with different complexity and performance trade-offs. Specifically, SN-Net splits a family of pretrained neural networks, which we call anchors, across the blocks/layers and then stitches them together with simple stitching layers to map the activations from one anchor to another. With only a few epochs of training, SN-Net effectively interpolates between the performance of anchors with varying scales. At runtime, SN-Net can instantly adapt to dynamic resource constraints by switching the stitching positions. Furthermore, we provide a comprehensive study on what, how and where to stitch as well as a simple strategy for effectively and efficiently training SN-Net. Extensive experiments on ImageNet classification demonstrate that SN-Net can obtain on-par or even better performance than many individually trained networks while supporting diverse deployment scenarios. For example, by stitching Swin Transformers, we challenge hundreds of models in Timm model zoo with a single network. We believe this new elastic model framework can serve as a strong baseline for further research in wider communities.