【深度学习】注意力机制(七)Agent Attention

本文介绍Agent Attention注意力机制,Transformer中的Attention模块可以提取全局语义信息,但是计算量太大,Agent Attention是一种计算非常有效的Attention模块。

论文:Agent Attention: On the Integration of Softmax and Linear Attention

代码:https://github.comA/leaplabthu/agent-attention

一、模块结构

Softmax Attention,Linear Attention, Agent Attention结构如下图:

Softmax Attention先进行Q和K的矩阵乘法,然后经过softmax并与V相乘,计算量大。

Linear Attention先进行K和V的矩阵乘法,然后再与Q相乘,降低了计算量。

Agent Attention引入了agent token A,A的维度为(n,d),n远小于N,通过与A的矩阵乘法降低了Q,K的维度,进而降低计算量。

二、推理公式

传统Attention的计算如下(x是输入,W是权重):

Softmax Attention就是将上式中的Sim(Q,K)变成了下式:

而Linear Attention的Sim(Q,K)如下:

为了简单起见,可以将Softmax Attention和Linear Attention写成下式:

那么Agent Attention可以写成:

等价于下式(A是引入的Agent token):

下图是Agent Attention的示意图(可以看到与最上面的图和上式很相似):

三、代码

Agent Attention可以放入各种Transformer模块中,这里展示PVT中使用Agent Attention的代码(就是将PVT原有的Attention模块替换成Agent Attention):

# -----------------------------------------------------------------------
# Agent Attention: On the Integration of Softmax and Linear Attention
# Modified by Dongchen Han
# -----------------------------------------------------------------------import torch
import torch.nn as nn
import torch.nn.functional as F
from functools import partialfrom timm.models.layers import DropPath, to_2tuple, trunc_normal_
from timm.models.registry import register_model
from timm.models.vision_transformer import _cfg__all__ = ['agent_pvt_tiny', 'agent_pvt_small', 'agent_pvt_medium', 'agent_pvt_large'
]class Mlp(nn.Module):def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=nn.GELU, drop=0.):super().__init__()out_features = out_features or in_featureshidden_features = hidden_features or in_featuresself.fc1 = nn.Linear(in_features, hidden_features)self.act = act_layer()self.fc2 = nn.Linear(hidden_features, out_features)self.drop = nn.Dropout(drop)def forward(self, x):x = self.fc1(x)x = self.act(x)x = self.drop(x)x = self.fc2(x)x = self.drop(x)return xclass Attention(nn.Module):def __init__(self, dim, num_heads=8, qkv_bias=False, qk_scale=None, attn_drop=0., proj_drop=0., sr_ratio=1):super().__init__()assert dim % num_heads == 0, f"dim {dim} should be divided by num_heads {num_heads}."self.dim = dimself.num_heads = num_headshead_dim = dim // num_headsself.scale = qk_scale or head_dim ** -0.5self.q = nn.Linear(dim, dim, bias=qkv_bias)self.kv = nn.Linear(dim, dim * 2, bias=qkv_bias)self.attn_drop = nn.Dropout(attn_drop)self.proj = nn.Linear(dim, dim)self.proj_drop = nn.Dropout(proj_drop)self.sr_ratio = sr_ratioif sr_ratio > 1:self.sr = nn.Conv2d(dim, dim, kernel_size=sr_ratio, stride=sr_ratio)self.norm = nn.LayerNorm(dim)def forward(self, x, H, W):B, N, C = x.shapeq = self.q(x).reshape(B, N, self.num_heads, C // self.num_heads).permute(0, 2, 1, 3)if self.sr_ratio > 1:x_ = x.permute(0, 2, 1).reshape(B, C, H, W)x_ = self.sr(x_).reshape(B, C, -1).permute(0, 2, 1)x_ = self.norm(x_)kv = self.kv(x_).reshape(B, -1, 2, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)else:kv = self.kv(x).reshape(B, -1, 2, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)k, v = kv[0], kv[1]attn = (q @ k.transpose(-2, -1)) * self.scaleattn = attn.softmax(dim=-1)attn = self.attn_drop(attn)x = (attn @ v).transpose(1, 2).reshape(B, N, C)x = self.proj(x)x = self.proj_drop(x)return xclass AgentAttention(nn.Module):def __init__(self, dim, num_patches, num_heads=8, qkv_bias=False, qk_scale=None, attn_drop=0., proj_drop=0.,sr_ratio=1, agent_num=49, **kwargs):super().__init__()assert dim % num_heads == 0, f"dim {dim} should be divided by num_heads {num_heads}."self.dim = dimself.num_patches = num_patcheswindow_size = (int(num_patches ** 0.5), int(num_patches ** 0.5))self.window_size = window_sizeself.num_heads = num_headshead_dim = dim // num_headsself.scale = head_dim ** -0.5self.q = nn.Linear(dim, dim, bias=qkv_bias)self.kv = nn.Linear(dim, dim * 2, bias=qkv_bias)self.attn_drop = nn.Dropout(attn_drop)self.proj = nn.Linear(dim, dim)self.proj_drop = nn.Dropout(proj_drop)self.sr_ratio = sr_ratioif sr_ratio > 1:self.sr = nn.Conv2d(dim, dim, kernel_size=sr_ratio, stride=sr_ratio)self.norm = nn.LayerNorm(dim)self.agent_num = agent_numself.dwc = nn.Conv2d(in_channels=dim, out_channels=dim, kernel_size=(3, 3), padding=1, groups=dim)self.an_bias = nn.Parameter(torch.zeros(num_heads, agent_num, 7, 7))self.na_bias = nn.Parameter(torch.zeros(num_heads, agent_num, 7, 7))self.ah_bias = nn.Parameter(torch.zeros(1, num_heads, agent_num, window_size[0] // sr_ratio, 1))self.aw_bias = nn.Parameter(torch.zeros(1, num_heads, agent_num, 1, window_size[1] // sr_ratio))self.ha_bias = nn.Parameter(torch.zeros(1, num_heads, window_size[0], 1, agent_num))self.wa_bias = nn.Parameter(torch.zeros(1, num_heads, 1, window_size[1], agent_num))trunc_normal_(self.an_bias, std=.02)trunc_normal_(self.na_bias, std=.02)trunc_normal_(self.ah_bias, std=.02)trunc_normal_(self.aw_bias, std=.02)trunc_normal_(self.ha_bias, std=.02)trunc_normal_(self.wa_bias, std=.02)pool_size = int(agent_num ** 0.5)self.pool = nn.AdaptiveAvgPool2d(output_size=(pool_size, pool_size))self.softmax = nn.Softmax(dim=-1)def forward(self, x, H, W):b, n, c = x.shapenum_heads = self.num_headshead_dim = c // num_headsq = self.q(x)if self.sr_ratio > 1:x_ = x.permute(0, 2, 1).reshape(b, c, H, W)x_ = self.sr(x_).reshape(b, c, -1).permute(0, 2, 1)x_ = self.norm(x_)kv = self.kv(x_).reshape(b, -1, 2, c).permute(2, 0, 1, 3)else:kv = self.kv(x).reshape(b, -1, 2, c).permute(2, 0, 1, 3)k, v = kv[0], kv[1]agent_tokens = self.pool(q.reshape(b, H, W, c).permute(0, 3, 1, 2)).reshape(b, c, -1).permute(0, 2, 1)q = q.reshape(b, n, num_heads, head_dim).permute(0, 2, 1, 3)k = k.reshape(b, n, num_heads, head_dim).permute(0, 2, 1, 3)v = v.reshape(b, n, num_heads, head_dim).permute(0, 2, 1, 3)agent_tokens = agent_tokens.reshape(b, self.agent_num, num_heads, head_dim).permute(0, 2, 1, 3)kv_size = (self.window_size[0] // self.sr_ratio, self.window_size[1] // self.sr_ratio)position_bias1 = nn.functional.interpolate(self.an_bias, size=kv_size, mode='bilinear')position_bias1 = position_bias1.reshape(1, num_heads, self.agent_num, -1).repeat(b, 1, 1, 1)position_bias2 = (self.ah_bias + self.aw_bias).reshape(1, num_heads, self.agent_num, -1).repeat(b, 1, 1, 1)position_bias = position_bias1 + position_bias2agent_attn = self.softmax((agent_tokens * self.scale) @ k.transpose(-2, -1) + position_bias)agent_attn = self.attn_drop(agent_attn)agent_v = agent_attn @ vagent_bias1 = nn.functional.interpolate(self.na_bias, size=self.window_size, mode='bilinear')agent_bias1 = agent_bias1.reshape(1, num_heads, self.agent_num, -1).permute(0, 1, 3, 2).repeat(b, 1, 1, 1)agent_bias2 = (self.ha_bias + self.wa_bias).reshape(1, num_heads, -1, self.agent_num).repeat(b, 1, 1, 1)agent_bias = agent_bias1 + agent_bias2q_attn = self.softmax((q * self.scale) @ agent_tokens.transpose(-2, -1) + agent_bias)q_attn = self.attn_drop(q_attn)x = q_attn @ agent_vx = x.transpose(1, 2).reshape(b, n, c)v = v.transpose(1, 2).reshape(b, H, W, c).permute(0, 3, 1, 2)x = x + self.dwc(v).permute(0, 2, 3, 1).reshape(b, n, c)x = self.proj(x)x = self.proj_drop(x)return xclass Block(nn.Module):def __init__(self, dim, num_patches, num_heads, mlp_ratio=4., qkv_bias=False, qk_scale=None, drop=0., attn_drop=0.,drop_path=0., act_layer=nn.GELU, norm_layer=nn.LayerNorm, sr_ratio=1,agent_num=49, attn_type='A'):super().__init__()self.norm1 = norm_layer(dim)assert attn_type in ['A', 'B']if attn_type == 'A':self.attn = AgentAttention(dim, num_patches,num_heads=num_heads, qkv_bias=qkv_bias, qk_scale=qk_scale,attn_drop=attn_drop, proj_drop=drop, sr_ratio=sr_ratio,agent_num=agent_num)else:self.attn = Attention(dim,num_heads=num_heads, qkv_bias=qkv_bias, qk_scale=qk_scale,attn_drop=attn_drop, proj_drop=drop, sr_ratio=sr_ratio)# NOTE: drop path for stochastic depth, we shall see if this is better than dropout hereself.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()self.norm2 = norm_layer(dim)mlp_hidden_dim = int(dim * mlp_ratio)self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)def forward(self, x, H, W):x = x + self.drop_path(self.attn(self.norm1(x), H, W))x = x + self.drop_path(self.mlp(self.norm2(x)))return xclass PatchEmbed(nn.Module):""" Image to Patch Embedding"""def __init__(self, img_size=224, patch_size=16, in_chans=3, embed_dim=768):super().__init__()img_size = to_2tuple(img_size)patch_size = to_2tuple(patch_size)self.img_size = img_sizeself.patch_size = patch_size# assert img_size[0] % patch_size[0] == 0 and img_size[1] % patch_size[1] == 0, \#     f"img_size {img_size} should be divided by patch_size {patch_size}."self.H, self.W = img_size[0] // patch_size[0], img_size[1] // patch_size[1]self.num_patches = self.H * self.Wself.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size, stride=patch_size)self.norm = nn.LayerNorm(embed_dim)def forward(self, x):B, C, H, W = x.shapex = self.proj(x).flatten(2).transpose(1, 2)x = self.norm(x)H, W = H // self.patch_size[0], W // self.patch_size[1]return x, (H, W)class PyramidVisionTransformer(nn.Module):def __init__(self, img_size=224, patch_size=16, in_chans=3, num_classes=1000, embed_dims=[64, 128, 256, 512],num_heads=[1, 2, 4, 8], mlp_ratios=[4, 4, 4, 4], qkv_bias=False, qk_scale=None, drop_rate=0.,attn_drop_rate=0., drop_path_rate=0., norm_layer=nn.LayerNorm,depths=[3, 4, 6, 3], sr_ratios=[8, 4, 2, 1], agent_sr_ratios='1111', num_stages=4,agent_num=[9, 16, 49, 49], attn_type='AAAA', **kwargs):super().__init__()self.num_classes = num_classesself.depths = depthsself.num_stages = num_stagesdpr = [x.item() for x in torch.linspace(0, drop_path_rate, sum(depths))]  # stochastic depth decay rulecur = 0attn_type = 'AAAA' if attn_type is None else attn_typefor i in range(num_stages):patch_embed = PatchEmbed(img_size=img_size if i == 0 else img_size // (2 ** (i - 1) * patch_size),patch_size=patch_size if i == 0 else 2,in_chans=in_chans if i == 0 else embed_dims[i - 1],embed_dim=embed_dims[i])num_patches = patch_embed.num_patchespos_embed = nn.Parameter(torch.zeros(1, num_patches, embed_dims[i]))pos_drop = nn.Dropout(p=drop_rate)block = nn.ModuleList([Block(dim=embed_dims[i], num_patches=num_patches, num_heads=num_heads[i], mlp_ratio=mlp_ratios[i], qkv_bias=qkv_bias,qk_scale=qk_scale, drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[cur + j],norm_layer=norm_layer, sr_ratio=sr_ratios[i] if attn_type[i] == 'B' else int(agent_sr_ratios[i]),agent_num=int(agent_num[i]), attn_type=attn_type[i])for j in range(depths[i])])cur += depths[i]setattr(self, f"patch_embed{i + 1}", patch_embed)setattr(self, f"pos_embed{i + 1}", pos_embed)setattr(self, f"pos_drop{i + 1}", pos_drop)setattr(self, f"block{i + 1}", block)# classification headself.head = nn.Linear(embed_dims[3], num_classes) if num_classes > 0 else nn.Identity()# init weightsfor i in range(num_stages):pos_embed = getattr(self, f"pos_embed{i + 1}")trunc_normal_(pos_embed, std=.02)# trunc_normal_(self.cls_token, std=.02)self.apply(self._init_weights)def _init_weights(self, m):if isinstance(m, nn.Linear):trunc_normal_(m.weight, std=.02)if isinstance(m, nn.Linear) and m.bias is not None:nn.init.constant_(m.bias, 0)elif isinstance(m, nn.LayerNorm):nn.init.constant_(m.bias, 0)nn.init.constant_(m.weight, 1.0)@torch.jit.ignoredef no_weight_decay(self):# return {'pos_embed', 'cls_token'} # has pos_embed may be betterreturn {'cls_token'}def get_classifier(self):return self.headdef reset_classifier(self, num_classes, global_pool=''):self.num_classes = num_classesself.head = nn.Linear(self.embed_dim, num_classes) if num_classes > 0 else nn.Identity()def _get_pos_embed(self, pos_embed, patch_embed, H, W):if H * W == self.patch_embed1.num_patches:return pos_embedelse:return F.interpolate(pos_embed.reshape(1, patch_embed.H, patch_embed.W, -1).permute(0, 3, 1, 2),size=(H, W), mode="bilinear").reshape(1, -1, H * W).permute(0, 2, 1)def forward_features(self, x):B = x.shape[0]for i in range(self.num_stages):patch_embed = getattr(self, f"patch_embed{i + 1}")pos_embed = getattr(self, f"pos_embed{i + 1}")pos_drop = getattr(self, f"pos_drop{i + 1}")block = getattr(self, f"block{i + 1}")x, (H, W) = patch_embed(x)pos_embed = self._get_pos_embed(pos_embed, patch_embed, H, W)x = pos_drop(x + pos_embed)for blk in block:x = blk(x, H, W)if i != self.num_stages - 1:x = x.reshape(B, H, W, -1).permute(0, 3, 1, 2).contiguous()return x.mean(dim=1)def forward(self, x):x = self.forward_features(x)x = self.head(x)return xdef _conv_filter(state_dict, patch_size=16):""" convert patch embedding weight from manual patchify + linear proj to conv"""out_dict = {}for k, v in state_dict.items():if 'patch_embed.proj.weight' in k:v = v.reshape((v.shape[0], 3, patch_size, patch_size))out_dict[k] = vreturn out_dictdef agent_pvt_tiny(pretrained=False, **kwargs):model = PyramidVisionTransformer(patch_size=4, embed_dims=[64, 128, 320, 512], num_heads=[1, 2, 5, 8], mlp_ratios=[8, 8, 4, 4], qkv_bias=True,norm_layer=partial(nn.LayerNorm, eps=1e-6), depths=[2, 2, 2, 2], sr_ratios=[8, 4, 2, 1],**kwargs)model.default_cfg = _cfg()return modeldef agent_pvt_small(pretrained=False, **kwargs):model = PyramidVisionTransformer(patch_size=4, embed_dims=[64, 128, 320, 512], num_heads=[1, 2, 5, 8], mlp_ratios=[8, 8, 4, 4], qkv_bias=True,norm_layer=partial(nn.LayerNorm, eps=1e-6), depths=[3, 4, 6, 3], sr_ratios=[8, 4, 2, 1], **kwargs)model.default_cfg = _cfg()return modeldef agent_pvt_medium(pretrained=False, **kwargs):model = PyramidVisionTransformer(patch_size=4, embed_dims=[64, 128, 320, 512], num_heads=[1, 2, 5, 8], mlp_ratios=[8, 8, 4, 4], qkv_bias=True,norm_layer=partial(nn.LayerNorm, eps=1e-6), depths=[3, 4, 18, 3], sr_ratios=[8, 4, 2, 1],**kwargs)model.default_cfg = _cfg()return modeldef agent_pvt_large(pretrained=False, **kwargs):model = PyramidVisionTransformer(patch_size=4, embed_dims=[64, 128, 320, 512], num_heads=[1, 2, 5, 8], mlp_ratios=[8, 8, 4, 4], qkv_bias=True,norm_layer=partial(nn.LayerNorm, eps=1e-6), depths=[3, 8, 27, 3], sr_ratios=[8, 4, 2, 1],**kwargs)model.default_cfg = _cfg()return model

本文来自互联网用户投稿,该文观点仅代表作者本人,不代表本站立场。本站仅提供信息存储空间服务,不拥有所有权,不承担相关法律责任。如若转载,请注明出处:http://www.mzph.cn/news/237284.shtml

如若内容造成侵权/违法违规/事实不符,请联系多彩编程网进行投诉反馈email:809451989@qq.com,一经查实,立即删除!

相关文章

前端手动部署与自动化部署

连接服务器 先购买服务器 安装vscode插件 连接服务器 连接成功 手动部署 安装nginx 启动nginx systemctl start nginx systemctl status nginx systemctl enable nginx启动 检查状态 开机就启动nginx 开始手动部署 配置nginx 成功

走过的2023:在挑战中领悟,在仿徨中成长

转眼间就到了2023年的最后一个月,回顾这短暂而又有意义的一年,可以用12个字总结:在挑战中领悟,在仿徨中成长。这篇文章我会从技术成长、职场生活、读书感悟和个人生活等几个方面,总结一下过去的这一年,梳理…

【玩转TableAgent数据智能分析】——个人体验分享

文章目录 前言上手体验优势不足 再次体验第三次体验第四次体验1、找到高价房源和低价房源的特点,看清民宿行业的整体布局2、了解各个地域的整体价格,优选潜力城市3、对比各个城市的评分,深入了解不同城市的民宿市场特点4、对比不同床型价格&a…

Java AQS 阻塞式锁和相关同步器工具的框架

8 J.U.C Java 并发工具包 AQS 原理 AQS:AbstractQueuedSynchronizer(抽象队列同步器),阻塞式锁和相关同步器工具的框架 特点: 用 state 属性来表示资源的状态(分独占模式和共享模式)&#…

TCP/IP 传输层协议

传输层定义了主机应用程序之间端到端的连通性。传输层中最为常见的两个协议分别是传输控制协议TCP(Transmission Control Protocol)和用户数据包协议UDP(User Datagram Protocol)。 TCP协议 TCP是一种面向连接的传输层协议&#…

esp32-s3解决使用蓝牙ble一键配网时,蓝牙ble内存使用的内部空间,空间不足时可采用外部PSRAM

idf.py menuconfig进入到esp32配置界面,配置NimBLE使用外部PSRAM内存即可

pip 离线安装:利用pypi网站进行模块 库的离线安装

离线安装是一种很好的方法,在网络不佳、库版本不明确、复杂库本地编译安装报错时,通过whl文件的下载安装,可以很高效的解决问题。 pypi的网站:https://pypi.org/ 这个网站包含各种你 pip install xxx 的库,离线安装可…

我做了一个在手机灵动岛锁屏看实时网速/步数/下班倒计时/跑步距离/照片/待办/倒计时/手机使用次数/帧率...的软件

我做了一个在手机灵动岛&锁屏看实时网速/步数/下班倒计时/跑步距离/照片/待办/倒计时/手机使用次数/帧率…的软件 Island Widgets 的作用: 提醒您 : 准时下班每天运动陪伴家人保持体重放下手机每日待办当前网速手机使用强度实时热搜现在天气… 初…

Python库学习(十三):爬虫框架Scrapy

微信搜索【猿码记】查看更多文章... 1.介绍 Scrapy是一个用于爬取网站数据的Python框架。它提供了一套强大而灵活的工具,使开发者能够轻松地创建和管理爬虫,从而从网站中提取所需的信息。框架要求Python的版本 3.8 Github Star:49.6k: https://github.c…

【Linux】Linux线程概念和线程控制

文章目录 一、Linux线程概念1.什么是线程2.线程的优缺点3.线程异常4.线程用途5.Linux进程VS线程 二、线程控制1.线程创建2.线程终止3.线程等待4.线程分离 一、Linux线程概念 1.什么是线程 线程是进程内的一个执行流。 我们知道,一个进程会有对应的PCB,…

GEE-Sentinel-2月度时间序列数据合成并导出

系列文章目录 第一章:时间序列数据合成 文章目录 系列文章目录前言时间序列数据合成总结 前言 利用每个月可获取植被指数数据取均值,合成月度平均植被指数,然后将12个月中的数据合成一个12波段的时间数据合成数据。 时间序列数据合成 代码…

ROS2 学习09--ros 中的通信接口的定义以及如何创建自定义msg、srv和action文件

在ROS系统中,无论话题还是服务,或者我们后续将要学习的动作,都会用到一个重要的概念——通信接口。 通信并不是一个人自言自语,而是两个甚至更多个人,你来我往的交流,交流的内容是什么呢?为了让…

数值分析(只为应付考试)

概述 研一时为应付高等工程数学考试整理的有关数值分析部分的内容,目的是为了应付考试。 误差 误差限与有效数字的联系 对于有 n n n 位有效数字的 x x x 的近似值 x ∗ x^* x∗, 其科学计数法表示形式 x ∗ a 1 . a 2 . . . a n 1 0 m ( a 1 ≠ 0 ) x^* …

Python开发GUI常用库PyQt6和PySide6介绍之二:设计师(Designer)

Python开发GUI常用库PyQt6和PySide6介绍之二:设计师(Designer) PySide6和PyQt6都有自己的设计师(Designer),用于可视化地设计和布局GUI应用程序的界面。这些设计师提供了丰富的工具和功能,使开…

文件名生成excel清单,怎么操作?这里有简单办法

文件名生成excel清单,怎么操作?为了整理文件名称,有时候需要将所有的文件名称整理好并且生成excel清单,大家可能还不能理解是什么意思,其实就是将所有文件的名称整理到excel表格里,形成一个清单。这个操作很…

【ECharts】雷达图

let chart echarts.init(this.$refs.radar_chart); let option {title: {text: 关键过程指标,},grid: {left: 0,},legend: {data: [个人, 小组, 团队],bottom: 0,itemWidth: 6,itemHeight: 6,},radar: {// shape: circle,indicator: [{ name: 成交额, max: 30000 },{ name: 成…

yum install net-tools 命令报错,无法安装成功

编辑网卡文件 插入数据,输入: i 保存编辑:输入 Esc 然后:wq

数据结构学习 leetcode64最小路径和

动态规划 题目: 建议看这里,有这道题详细的解析。我觉得写的挺好。 这是我在学动态规划的时候,动手做的一道题。 虽然我在学动态规划,但是我之前学了dps,所以我就想先用dps试着做,结果发现不行&#xf…

使用Gitee中的CI/CD来完成代码的自动部署与发布(使用内网穿透把本地电脑当作服务器使用)

📚目录 📚简介:⚙️ 所需工具:💨内网穿透配置💭工具介绍✨命令安装🎊配置Cpolar🕳️关闭防火墙🥛防火墙端口放行规则(关闭防火墙可以忽略)🍬小章总…

打造明厨亮灶工程,需要哪些AI视频智能算法助力?

旭帆科技AI智能监控可以通过摄像头、传感器和数据处理等技术手段,实时监测厨房人员着装、行为与烟火等,对厨房实时监控进行分析与记录,从而实现明厨亮灶场景的搭建,保障食品安全和服务质量。 1、烟火识别 对于后厨来说&#xff0…