CycleMLP: A MLP-like Architecture for Dense Prediction
This work addresses a bottleneck in applying MLP architectures to computer vision tasks like object detection and segmentation, offering a more efficient and versatile backbone.
The paper tackles the problem of MLP-like architectures being infeasible for dense prediction tasks due to computational inefficiency and fixed image size constraints, resulting in CycleMLP, which achieves linear computational complexity and outperforms state-of-the-art models like Swin Transformer with fewer parameters and FLOPs, e.g., a 1.3% mIoU improvement on ADE20K.
This paper presents a simple MLP-like architecture, CycleMLP, which is a versatile backbone for visual recognition and dense predictions. As compared to modern MLP architectures, e.g., MLP-Mixer, ResMLP, and gMLP, whose architectures are correlated to image size and thus are infeasible in object detection and segmentation, CycleMLP has two advantages compared to modern approaches. (1) It can cope with various image sizes. (2) It achieves linear computational complexity to image size by using local windows. In contrast, previous MLPs have $O(N^2)$ computations due to fully spatial connections. We build a family of models which surpass existing MLPs and even state-of-the-art Transformer-based models, e.g., Swin Transformer, while using fewer parameters and FLOPs. We expand the MLP-like models' applicability, making them a versatile backbone for dense prediction tasks. CycleMLP achieves competitive results on object detection, instance segmentation, and semantic segmentation. In particular, CycleMLP-Tiny outperforms Swin-Tiny by 1.3% mIoU on ADE20K dataset with fewer FLOPs. Moreover, CycleMLP also shows excellent zero-shot robustness on ImageNet-C dataset. Code is available at https://github.com/ShoufaChen/CycleMLP.