Home Technology peripherals AI The visual representation model vHeat inspired by physical heat transfer is here. It attempts to break through the attention mechanism and has both low complexity and global receptive field.

The visual representation model vHeat inspired by physical heat transfer is here. It attempts to break through the attention mechanism and has both low complexity and global receptive field.

Jun 06, 2024 pm 05:28 PM
industry vHeat visual representation model

The visual representation model vHeat inspired by physical heat transfer is here. It attempts to break through the attention mechanism and has both low complexity and global receptive field.
The AIxiv column is a column where this site publishes academic and technical content. In the past few years, the AIxiv column of this site has received more than 2,000 reports, covering top laboratories from major universities and companies around the world, effectively promoting academic exchanges and dissemination. If you have excellent work that you want to share, please feel free to contribute or contact us for reporting. Submission email: liyazhou@jiqizhixin.com; zhaoyunfeng@jiqizhixin.com

##The author of this article is the original team of VMamba, among whom the first The author Wang Zhaozhi is a 2022 doctoral student jointly trained by the University of Chinese Academy of Sciences and Pengcheng Laboratory, and the co-author Liu Yue is a 2021 direct doctoral student of the University of Chinese Academy of Sciences. Their main research directions are visual model design and self-supervised learning.
How to break through the Attention mechanism of Transformer? The University of Chinese Academy of Sciences and Pengcheng National Laboratory proposed a visual representation model vHeat

based on heat conduction. Treat image feature blocks as heat sources, and extract image features by predicting thermal conductivity and using physical heat conductionprinciples. Compared with visual models based on the Attention mechanism, vHeat takes into account: computational complexity (1.5th power), global receptive field, and physical interpretability. When using vHeat-base+%E6%A8%A1%E5%9E%8B for high-resolution image input, the put, GPU memory usage, and flops are Swin-base+%E6%A8%A1 respectively. 3 times, 1/4, 3/4 of %E5%9E%8B. It has achieved advanced performance on basic downstream tasks such as image classification, target detection, and semantic/instance segmentation.

The visual representation model vHeat inspired by physical heat transfer is here. It attempts to break through the attention mechanism and has both low complexity and global receptive field.

    Paper address: https://arxiv.org/pdf/2405.16555
  • Code address: https://github.com/MzeroMiko/vHeat
  • Paper title: vHeat: Building Vision Models upon Heat Conduction
Overview

The two most mainstream basic visual models currently are CNN and Visual Transformer (ViT). However, the performance of CNN is limited by local receptive fields and fixed convolution kernel operators. ViT has the ability to represent global dependencies, but at the cost of high quadratic norm computational complexity. We believe that the convolution operators and self-attention operators of CNN and ViT are both pixel transfer processes within features, which are respectively a form of information transfer, which also reminds us of heat conduction in the physical field. So based on the heat conduction equation, we connected the spatial propagation of visual semantics with physical heat conduction, proposed a visual conduction operator (Heat Conduction Operator, HCO) with 1.5 power computational complexity, and then designed a heat conduction operator with low Visual representation model vHeat for complexity, global receptive field, and physical interpretability. The calculation form and complexity comparison between HCO and self-attention are shown in the figure below. Experiments have proven that vHeat performs well in various visual tasks. For example, vHeat-T achieves 82.2% classification accuracy on ImageNet-1K, which is 0.9% higher than Swin-T and 1.7% higher than ViM-S. In addition to performance, vHeat also has the advantages of high inference speed, low GPU memory usage and low FLOPs. When the input image resolution is high, the base-scale vHeat model only has 1/3 more throughput, 1/4 GPU memory usage, and 3/4 FLOPs compared to Swin.

The visual representation model vHeat inspired by physical heat transfer is here. It attempts to break through the attention mechanism and has both low complexity and global receptive field.

Method introduction

Use

to represent the temperature of point

at time t, and the physical heat conduction equation is The visual representation model vHeat inspired by physical heat transfer is here. It attempts to break through the attention mechanism and has both low complexity and global receptive field., where k>0, represents thermal diffusivity. Given the initial conditionsThe visual representation model vHeat inspired by physical heat transfer is here. It attempts to break through the attention mechanism and has both low complexity and global receptive field. at time t=0, the heat conduction equation can be solved by using Fourier transform, which is expressed as follows:The visual representation model vHeat inspired by physical heat transfer is here. It attempts to break through the attention mechanism and has both low complexity and global receptive field.

The visual representation model vHeat inspired by physical heat transfer is here. It attempts to break through the attention mechanism and has both low complexity and global receptive field.

where The visual representation model vHeat inspired by physical heat transfer is here. It attempts to break through the attention mechanism and has both low complexity and global receptive field. and The visual representation model vHeat inspired by physical heat transfer is here. It attempts to break through the attention mechanism and has both low complexity and global receptive field. represent Fourier transform and inverse Fourier transform respectively, The visual representation model vHeat inspired by physical heat transfer is here. It attempts to break through the attention mechanism and has both low complexity and global receptive field. represents frequency domain spatial coordinates.

We use HCO to implement heat conduction in visual semantics. First, we extend the The visual representation model vHeat inspired by physical heat transfer is here. It attempts to break through the attention mechanism and has both low complexity and global receptive field. in the physical heat conduction equation to a multi-channel feature The visual representation model vHeat inspired by physical heat transfer is here. It attempts to break through the attention mechanism and has both low complexity and global receptive field., and treat The visual representation model vHeat inspired by physical heat transfer is here. It attempts to break through the attention mechanism and has both low complexity and global receptive field. as input and The visual representation model vHeat inspired by physical heat transfer is here. It attempts to break through the attention mechanism and has both low complexity and global receptive field. as output. , HCO simulates the general solution of heat conduction in the discretized form, as shown in the following formula:

The visual representation model vHeat inspired by physical heat transfer is here. It attempts to break through the attention mechanism and has both low complexity and global receptive field.

where The visual representation model vHeat inspired by physical heat transfer is here. It attempts to break through the attention mechanism and has both low complexity and global receptive field. and The visual representation model vHeat inspired by physical heat transfer is here. It attempts to break through the attention mechanism and has both low complexity and global receptive field. represent the two-dimensional discrete cosine transform and the inverse transform respectively, The structure of HCO is shown in Figure (a) below.

The visual representation model vHeat inspired by physical heat transfer is here. It attempts to break through the attention mechanism and has both low complexity and global receptive field.

In addition, we believe that different image contents should correspond to different thermal diffusivities. Considering that the output of The visual representation model vHeat inspired by physical heat transfer is here. It attempts to break through the attention mechanism and has both low complexity and global receptive field. is in the frequency domain, we determine the thermal diffusion based on the frequency value Rate,The visual representation model vHeat inspired by physical heat transfer is here. It attempts to break through the attention mechanism and has both low complexity and global receptive field.. Since different positions in the frequency domain represent different frequency values, we propose Frequency Value Embeddings (FVEs) to represent frequency value information, which is similar to the implementation and function of absolute position encoding in ViT, and use FVEs to control heat diffusion. The rate k is predicted so that HCO can perform non-uniform and adaptive conduction, as shown in the figure below.

The visual representation model vHeat inspired by physical heat transfer is here. It attempts to break through the attention mechanism and has both low complexity and global receptive field.

vHeat is implemented using a multi-level structure, as shown in the figure below. The overall framework is similar to the mainstream visual model, and the HCO layer is shown in Figure 2 (b).

The visual representation model vHeat inspired by physical heat transfer is here. It attempts to break through the attention mechanism and has both low complexity and global receptive field.

Experimental results

ImageNet classification

The visual representation model vHeat inspired by physical heat transfer is here. It attempts to break through the attention mechanism and has both low complexity and global receptive field.

By comparing the experimental results, it is easy to see that under similar parameter amounts and FLOPs:

  1. vHeat-T achieved 82.2 The performance of % exceeds DeiT-S by 2.4%, Vim-S by 1.7%, and Swin-T by 0.9%.
  2. vHeat-S achieved 83.6% performance, exceeding Swin-S by 0.6% and ConvNeXt-S by 0.5%.
  3. vHeat-B achieved 83.9% performance, exceeding DeiT-B by 2.1% and Swin-B by 0.4%.

At the same time, due to vHeat's O (N^1.5) low complexity and parallel computation, the inference throughput is compared to ViTs and SSM The model has obvious advantages. For example, the inference throughput of vHeat-T is 1514 img/s, which is 22% higher than Swin-T and 87% higher than Vim-S. , it is also 26% higher than ConvNeXt-T, and has better performance.

Downstream tasks

The visual representation model vHeat inspired by physical heat transfer is here. It attempts to break through the attention mechanism and has both low complexity and global receptive field.

On the COCO data set, vHeat also has a performance advantage: in the case of fine-tune 12 epochs , vHeat-T/S/B reaches 45.1/46.8/47.7 mAP respectively, exceeding Swin-T/S/B by 2.4/2.0/0.8 mAP, and exceeding ConvNeXt-T/S/B by 0.9/1.4/0.7 mAP. On the ADE20K data set, vHeat-T/S/B reached 46.9/49.0/49.6 mIoU respectively, which still has better performance than Swin and ConvNeXt. These results verify that vHeat fully works in visual downstream experiments, demonstrating the potential to replace mainstream basic visual models.

Analysis Experiment

Effective Feeling Field

The visual representation model vHeat inspired by physical heat transfer is here. It attempts to break through the attention mechanism and has both low complexity and global receptive field.

vHeat has the global effective feeling However, among the mainstream models for visual comparison, only DeiT and HiViT also have this feature. However, it is worth noting that the cost of DeiT and HiViT is square level complexity, while vHeat is 1.5 power level complexity.

Calculate the cost

The visual representation model vHeat inspired by physical heat transfer is here. It attempts to break through the attention mechanism and has both low complexity and global receptive field.

The above picture from left to right is vHeat-B and Comparison of inference throughput/GPU memory usage/computation FLOPs of ViT-based models at other base scales. It can be clearly seen that due to the computational complexity of O (N^1.5), vHeat has faster inference speed, lower memory usage and fewer FLOPs than the contrasting models, and when the image resolution is larger, , the advantages will be more obvious. When the input image is 768*768 resolution, the inference throughput of vHeat-B is about 3 times that of Swin-B, and the GPU memory usage is 74% lower than that of Swin-B. FLOPs are 28% lower than Swin-B. Comparison of the computational cost of vHeat and ViT-based models demonstrates its excellent potential in processing high-resolution images.

The above is the detailed content of The visual representation model vHeat inspired by physical heat transfer is here. It attempts to break through the attention mechanism and has both low complexity and global receptive field.. For more information, please follow other related articles on the PHP Chinese website!

Statement of this Website
The content of this article is voluntarily contributed by netizens, and the copyright belongs to the original author. This site does not assume corresponding legal responsibility. If you find any content suspected of plagiarism or infringement, please contact admin@php.cn

Hot AI Tools

Undresser.AI Undress

Undresser.AI Undress

AI-powered app for creating realistic nude photos

AI Clothes Remover

AI Clothes Remover

Online AI tool for removing clothes from photos.

Undress AI Tool

Undress AI Tool

Undress images for free

Clothoff.io

Clothoff.io

AI clothes remover

Video Face Swap

Video Face Swap

Swap faces in any video effortlessly with our completely free AI face swap tool!

Hot Article

Roblox: Bubble Gum Simulator Infinity - How To Get And Use Royal Keys
3 weeks ago By 尊渡假赌尊渡假赌尊渡假赌
Nordhold: Fusion System, Explained
3 weeks ago By 尊渡假赌尊渡假赌尊渡假赌

Hot Tools

Notepad++7.3.1

Notepad++7.3.1

Easy-to-use and free code editor

SublimeText3 Chinese version

SublimeText3 Chinese version

Chinese version, very easy to use

Zend Studio 13.0.1

Zend Studio 13.0.1

Powerful PHP integrated development environment

Dreamweaver CS6

Dreamweaver CS6

Visual web development tools

SublimeText3 Mac version

SublimeText3 Mac version

God-level code editing software (SublimeText3)

Hot Topics

Java Tutorial
1664
14
PHP Tutorial
1268
29
C# Tutorial
1248
24
DeepMind robot plays table tennis, and its forehand and backhand slip into the air, completely defeating human beginners DeepMind robot plays table tennis, and its forehand and backhand slip into the air, completely defeating human beginners Aug 09, 2024 pm 04:01 PM

But maybe he can’t defeat the old man in the park? The Paris Olympic Games are in full swing, and table tennis has attracted much attention. At the same time, robots have also made new breakthroughs in playing table tennis. Just now, DeepMind proposed the first learning robot agent that can reach the level of human amateur players in competitive table tennis. Paper address: https://arxiv.org/pdf/2408.03906 How good is the DeepMind robot at playing table tennis? Probably on par with human amateur players: both forehand and backhand: the opponent uses a variety of playing styles, and the robot can also withstand: receiving serves with different spins: However, the intensity of the game does not seem to be as intense as the old man in the park. For robots, table tennis

The first mechanical claw! Yuanluobao appeared at the 2024 World Robot Conference and released the first chess robot that can enter the home The first mechanical claw! Yuanluobao appeared at the 2024 World Robot Conference and released the first chess robot that can enter the home Aug 21, 2024 pm 07:33 PM

On August 21, the 2024 World Robot Conference was grandly held in Beijing. SenseTime's home robot brand "Yuanluobot SenseRobot" has unveiled its entire family of products, and recently released the Yuanluobot AI chess-playing robot - Chess Professional Edition (hereinafter referred to as "Yuanluobot SenseRobot"), becoming the world's first A chess robot for the home. As the third chess-playing robot product of Yuanluobo, the new Guoxiang robot has undergone a large number of special technical upgrades and innovations in AI and engineering machinery. For the first time, it has realized the ability to pick up three-dimensional chess pieces through mechanical claws on a home robot, and perform human-machine Functions such as chess playing, everyone playing chess, notation review, etc.

Claude has become lazy too! Netizen: Learn to give yourself a holiday Claude has become lazy too! Netizen: Learn to give yourself a holiday Sep 02, 2024 pm 01:56 PM

The start of school is about to begin, and it’s not just the students who are about to start the new semester who should take care of themselves, but also the large AI models. Some time ago, Reddit was filled with netizens complaining that Claude was getting lazy. "Its level has dropped a lot, it often pauses, and even the output becomes very short. In the first week of release, it could translate a full 4-page document at once, but now it can't even output half a page!" https:// www.reddit.com/r/ClaudeAI/comments/1by8rw8/something_just_feels_wrong_with_claude_in_the/ in a post titled "Totally disappointed with Claude", full of

At the World Robot Conference, this domestic robot carrying 'the hope of future elderly care' was surrounded At the World Robot Conference, this domestic robot carrying 'the hope of future elderly care' was surrounded Aug 22, 2024 pm 10:35 PM

At the World Robot Conference being held in Beijing, the display of humanoid robots has become the absolute focus of the scene. At the Stardust Intelligent booth, the AI ​​robot assistant S1 performed three major performances of dulcimer, martial arts, and calligraphy in one exhibition area, capable of both literary and martial arts. , attracted a large number of professional audiences and media. The elegant playing on the elastic strings allows the S1 to demonstrate fine operation and absolute control with speed, strength and precision. CCTV News conducted a special report on the imitation learning and intelligent control behind "Calligraphy". Company founder Lai Jie explained that behind the silky movements, the hardware side pursues the best force control and the most human-like body indicators (speed, load) etc.), but on the AI ​​side, the real movement data of people is collected, allowing the robot to become stronger when it encounters a strong situation and learn to evolve quickly. And agile

ACL 2024 Awards Announced: One of the Best Papers on Oracle Deciphering by HuaTech, GloVe Time Test Award ACL 2024 Awards Announced: One of the Best Papers on Oracle Deciphering by HuaTech, GloVe Time Test Award Aug 15, 2024 pm 04:37 PM

At this ACL conference, contributors have gained a lot. The six-day ACL2024 is being held in Bangkok, Thailand. ACL is the top international conference in the field of computational linguistics and natural language processing. It is organized by the International Association for Computational Linguistics and is held annually. ACL has always ranked first in academic influence in the field of NLP, and it is also a CCF-A recommended conference. This year's ACL conference is the 62nd and has received more than 400 cutting-edge works in the field of NLP. Yesterday afternoon, the conference announced the best paper and other awards. This time, there are 7 Best Paper Awards (two unpublished), 1 Best Theme Paper Award, and 35 Outstanding Paper Awards. The conference also awarded 3 Resource Paper Awards (ResourceAward) and Social Impact Award (

Li Feifei's team proposed ReKep to give robots spatial intelligence and integrate GPT-4o Li Feifei's team proposed ReKep to give robots spatial intelligence and integrate GPT-4o Sep 03, 2024 pm 05:18 PM

Deep integration of vision and robot learning. When two robot hands work together smoothly to fold clothes, pour tea, and pack shoes, coupled with the 1X humanoid robot NEO that has been making headlines recently, you may have a feeling: we seem to be entering the age of robots. In fact, these silky movements are the product of advanced robotic technology + exquisite frame design + multi-modal large models. We know that useful robots often require complex and exquisite interactions with the environment, and the environment can be represented as constraints in the spatial and temporal domains. For example, if you want a robot to pour tea, the robot first needs to grasp the handle of the teapot and keep it upright without spilling the tea, then move it smoothly until the mouth of the pot is aligned with the mouth of the cup, and then tilt the teapot at a certain angle. . this

Distributed Artificial Intelligence Conference DAI 2024 Call for Papers: Agent Day, Richard Sutton, the father of reinforcement learning, will attend! Yan Shuicheng, Sergey Levine and DeepMind scientists will give keynote speeches Distributed Artificial Intelligence Conference DAI 2024 Call for Papers: Agent Day, Richard Sutton, the father of reinforcement learning, will attend! Yan Shuicheng, Sergey Levine and DeepMind scientists will give keynote speeches Aug 22, 2024 pm 08:02 PM

Conference Introduction With the rapid development of science and technology, artificial intelligence has become an important force in promoting social progress. In this era, we are fortunate to witness and participate in the innovation and application of Distributed Artificial Intelligence (DAI). Distributed artificial intelligence is an important branch of the field of artificial intelligence, which has attracted more and more attention in recent years. Agents based on large language models (LLM) have suddenly emerged. By combining the powerful language understanding and generation capabilities of large models, they have shown great potential in natural language interaction, knowledge reasoning, task planning, etc. AIAgent is taking over the big language model and has become a hot topic in the current AI circle. Au

Hongmeng Smart Travel S9 and full-scenario new product launch conference, a number of blockbuster new products were released together Hongmeng Smart Travel S9 and full-scenario new product launch conference, a number of blockbuster new products were released together Aug 08, 2024 am 07:02 AM

This afternoon, Hongmeng Zhixing officially welcomed new brands and new cars. On August 6, Huawei held the Hongmeng Smart Xingxing S9 and Huawei full-scenario new product launch conference, bringing the panoramic smart flagship sedan Xiangjie S9, the new M7Pro and Huawei novaFlip, MatePad Pro 12.2 inches, the new MatePad Air, Huawei Bisheng With many new all-scenario smart products including the laser printer X1 series, FreeBuds6i, WATCHFIT3 and smart screen S5Pro, from smart travel, smart office to smart wear, Huawei continues to build a full-scenario smart ecosystem to bring consumers a smart experience of the Internet of Everything. Hongmeng Zhixing: In-depth empowerment to promote the upgrading of the smart car industry Huawei joins hands with Chinese automotive industry partners to provide

See all articles