FFT-based Dynamic Token Mixer for Vision
This work addresses efficiency bottlenecks in computer vision models, offering a practical alternative for researchers and practitioners dealing with high-resolution images, though it appears incremental as it builds on existing token-mixer and MetaFormer architectures.
The paper tackles the high computational complexity of multi-head self-attention in vision models by proposing a novel FFT-based dynamic token mixer, resulting in improved throughput and memory efficiency for high-resolution image recognition tasks.
Multi-head-self-attention (MHSA)-equipped models have achieved notable performance in computer vision. Their computational complexity is proportional to quadratic numbers of pixels in input feature maps, resulting in slow processing, especially when dealing with high-resolution images. New types of token-mixer are proposed as an alternative to MHSA to circumvent this problem: an FFT-based token-mixer involves global operations similar to MHSA but with lower computational complexity. However, despite its attractive properties, the FFT-based token-mixer has not been carefully examined in terms of its compatibility with the rapidly evolving MetaFormer architecture. Here, we propose a novel token-mixer called Dynamic Filter and novel image recognition models, DFFormer and CDFFormer, to close the gaps above. The results of image classification and downstream tasks, analysis, and visualization show that our models are helpful. Notably, their throughput and memory efficiency when dealing with high-resolution image recognition is remarkable. Our results indicate that Dynamic Filter is one of the token-mixer options that should be seriously considered. The code is available at https://github.com/okojoalg/dfformer