MSG-Transformer: Exchanging Local Spatial Information by Manipulating Messenger Tokens
This work addresses efficiency bottlenecks for researchers and practitioners applying Transformers to visual recognition, though it is incremental as it builds on existing multi-scale architectures.
The paper tackles the computational inefficiency of Transformers in high-resolution visual tasks by introducing messenger tokens to exchange local spatial information, achieving competitive performance in image classification and object detection with accelerated inference on GPU and CPU.
Transformers have offered a new methodology of designing neural networks for visual recognition. Compared to convolutional networks, Transformers enjoy the ability of referring to global features at each stage, yet the attention module brings higher computational overhead that obstructs the application of Transformers to process high-resolution visual data. This paper aims to alleviate the conflict between efficiency and flexibility, for which we propose a specialized token for each region that serves as a messenger (MSG). Hence, by manipulating these MSG tokens, one can flexibly exchange visual information across regions and the computational complexity is reduced. We then integrate the MSG token into a multi-scale architecture named MSG-Transformer. In standard image classification and object detection, MSG-Transformer achieves competitive performance and the inference on both GPU and CPU is accelerated. Code is available at https://github.com/hustvl/MSG-Transformer.