论文阅读 [CVPR-2022] An Image Patch is a Wave: Quantum Inspired Vision MLP

论文阅读 [CVPR-2022] An Image Patch is a Wave: Quantum Inspired Vision MLP

studyai.com

搜索论文: An Image Patch is a Wave: Quantum Inspired Vision MLP

http://www.studyai.com/search/whole-site/?q=An+Image+Patch+is+a+Wave+Quantum+Inspired+Vision+MLP

摘要(Abstract)

In the field of computer vision, recent works show that a pure MLP architecture mainly stacked by fully-connected layers can achieve competing performance with CNN and transformer.

在计算机视觉领域，最近的研究表明，主要由全连接层堆叠的纯MLP架构可以实现与CNN和transformer相竞争的性能。

An input image of vision MLP is usually split into multiple tokens (patches), while the existing MLP models directly aggregate them with fixed weights, neglecting the varying semantic information of tokens from different images.

视觉MLP的输入图像通常被分割成多个标记（补丁），而现有的MLP模型直接使用固定权重对其进行聚合，忽略了不同图像中标记的不同语义信息。

To dynamically aggregate tokens, we propose to represent each token as a wave function with two parts, amplitude and phase.

为了动态聚合令牌，我们建议将每个令牌表示为一个包含两部分的波函数，振幅和相位。

Amplitude is the original feature and the phase term is a complex value changing according to the semantic contents of input images. Introducing the phase term can dynamically modulate the relationship between tokens and fixed weights in MLP.

振幅是原始特征，相位项是根据输入图像的语义内容变化的复值。引入相位项可以动态调整MLP中令牌和固定权重之间的关系。

Based on the wave-like token representation, we establish a novel Wave-MLP architecture for vision tasks.

基于wave-like-token表示，我们为视觉任务建立了一种新的wave-MLP体系结构。

Extensive experiments demonstrate that the proposed Wave-MLP is superior to the stateof-the-art MLP architectures on various vision tasks such as image classification, object detection and semantic segmentation.

大量实验表明，在图像分类、目标检测和语义分割等各种视觉任务上，所提出的Wave-MLP优于目前最先进的MLP体系结构。

The source code will be available at https: //github.com/huawei-noah/CV-Backbones/tree/master/wavemlp_pytorch and https://gitee.com/mindspore/models/tree/master/research/cv/wave_mlp.