Mobile-Former: Bridging MobileNet and Transformer
This work addresses the need for efficient and powerful vision models for mobile and edge devices, offering a novel architecture that outperforms existing methods in both classification and detection tasks.
The paper tackles the problem of combining local and global feature processing in vision tasks by introducing Mobile-Former, a parallel design of MobileNet and transformer with a two-way bridge, resulting in improved accuracy and efficiency; for example, it achieves 77.9% top-1 accuracy at 294M FLOPs on ImageNet, gaining 1.3% over MobileNetV3 while saving 17% computations.
We present Mobile-Former, a parallel design of MobileNet and transformer with a two-way bridge in between. This structure leverages the advantages of MobileNet at local processing and transformer at global interaction. And the bridge enables bidirectional fusion of local and global features. Different from recent works on vision transformer, the transformer in Mobile-Former contains very few tokens (e.g. 6 or fewer tokens) that are randomly initialized to learn global priors, resulting in low computational cost. Combining with the proposed light-weight cross attention to model the bridge, Mobile-Former is not only computationally efficient, but also has more representation power. It outperforms MobileNetV3 at low FLOP regime from 25M to 500M FLOPs on ImageNet classification. For instance, Mobile-Former achieves 77.9\% top-1 accuracy at 294M FLOPs, gaining 1.3\% over MobileNetV3 but saving 17\% of computations. When transferring to object detection, Mobile-Former outperforms MobileNetV3 by 8.6 AP in RetinaNet framework. Furthermore, we build an efficient end-to-end detector by replacing backbone, encoder and decoder in DETR with Mobile-Former, which outperforms DETR by 1.1 AP but saves 52\% of computational cost and 36\% of parameters.