self.norm为啥可以当函数调用_call_

原创 于 2025-01-17 11:19:48 发布 · 134 阅读 · 0 ·
CC 4.0 BY-SA版权
版权声明：本文为博主原创文章，遵循 CC 4.0 BY-SA 版权协议，转载请附上原文出处链接和本声明。
文章标签：

#python

8b2e29a9cbe2459fae62183da833cdb3.jpg

 

risc123456
关注
CrossNorm与SelfNorm的具体实现
qq_18943707的博客
03-27 332
主要功能：CrossNorm和SelfNorm模块通过自适应调整特征图的统计量，增强模型对分布偏移的鲁棒性。适用场景：在训练阶段使用这些模块可以提高模型泛化能力，而无需修改网络结构。
手写transformer
m0_53292725的博客
09-29 1094
【代码】手写transformer。
计算机视觉中的注意力机制
我爱计算机视觉
06-10 1972
本文转载自知乎，已获作者授权转载。https://zhuanlan.zhihu.com/p/146130215之前在看DETR这篇论文中的self_attention，然后结合之前实验室...
DETR代码学习笔记（二）
athrunsunny的博客
02-26 8029
上一篇DETR代码学习笔记（一）记录完了DETR的backbone以及数据进入encoder之前的部分，这篇开始介绍DETR的transformer部分，大体上DETR的transformer和Attention is All You Need中提出框架基本是一致的，只是数据上有些不同。 首先是encoder部分，整个encoder部分，从代码上可以很直观的看出他的整体流程，其中的最初的q，k是由backbone中得到的Feature Map加上位置编码得到，这个位置编.........
mmdetection源码解析--ResNet18
Recursions的博客
08-08 1169
mmdetection源码解析--ResNet18。
Traceback (most recent call last): [rank0]: File "/data3/workspace/chenzh/Face-SVD/train.py", line 1668, in <module> [rank0]: main(**OmegaConf.load(args.config)) [rank0]: File "/data3/workspace/chenzh/Face-SVD/train.py", line 1409, in main [rank0]: discr_fake_pred = attr_discriminator(vt_hat_latents) [rank0]: File "/data1/miniconda3/envs/face_svd/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1553, in _wrapped_call_impl [rank0]: return self._call_impl(*args, **kwargs) [rank0]: File "/data1/miniconda3/envs/face_svd/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1562, in _call_impl [rank0]: return forward_call(*args, **kwargs) [rank0]: File "/data1/miniconda3/envs/face_svd/lib/python3.10/site-packages/torch/nn/parallel/distributed.py", line 1632, in forward [rank0]: inputs, kwargs = self._pre_forward(*inputs, **kwargs) [rank0]: File "/data1/miniconda3/envs/face_svd/lib/python3.10/site-packages/torch/nn/parallel/distributed.py", line 1523, in _pre_forward [rank0]: if torch.is_grad_enabled() and self.reducer._rebuild_buckets(): [rank0]: RuntimeError: Expected to have finished reduction in the prior iteration before starting a new one. This error indicates that your module has parameters that were not used in producing loss. You can enable unused parameter detection by passing the keyword argument `find_unused_parameters=True` to `torch.nn.parallel.DistributedDataParallel`, and by [rank0]: making sure all `forward` function outputs participate in calculating loss. [rank0]: If you already have done the above, then the distributed data parallel module wasn't able to locate the output tensors in the return value of your module's `forward` function. Please include the loss function and the structure of the return value of `forward` of your module when reporting this issue (e.g. list, dict, iterable). [rank0]: Parameters which did not receive grad for rank 0: backbone.encoder_block3.0.norm.bias, backbone.encoder_block3.0.norm.weight, backbone.encoder_block3.0.conv.weight, backbone.encoder_block2.2.attn.to_out.0.bias, backbone.encoder_block2.2.attn.to_out.0.weight, backbone.encoder_block2.2.attn.to_v.weight, backbone.encoder_block2.2.attn.to_k.weight, backbone.encoder_block2.2.attn.to_q.weight, backbone.encoder_block2.2.norm.bias, backbone.encoder_block2.2.norm.weight, backbone.encoder_block2.2.resnet.conv2.bias, backbone.encoder_block2.2.resnet.conv2.weight, backbone.encoder_block2.2.resnet.norm2.bias, backbone.encoder_block2.2.resnet.norm2.weight, backbone.encoder_block2.2.resnet.conv1.bias, backbone.encoder_block2.2.resnet.conv1.weight, backbone.encoder_block2.2.resnet.norm1.bias, backbone.encoder_block2.2.resnet.norm1.weight, backbone.encoder_block2.1.attn.to_out.0.bias, backbone.encoder_block2.1.attn.to_out.0.weight, backbone.encoder_block2.1.attn.to_v.weight, backbone.encoder_block2.1.attn.to_k.weight, backbone.encoder_block2.1.attn.to_q.weight, backbone.encoder_block2.1.norm.bias, backbone.encoder_block2.1.norm.weight, backbone.encoder_block2.1.resnet.conv2.bias, backbone.encoder_block2.1.resnet.conv2.weight, backbone.encoder_block2.1.resnet.norm2.bias, backbone.encoder_block2.1.resnet.norm2.weight, backbone.encoder_block2.1.resnet.conv1.bias, backbone.encoder_block2.1.resnet.conv1.weight, backbone.encoder_block2.1.resnet.norm1.bias, backbone.encoder_block2.1.resnet.norm1.weight, backbone.encoder_block2.0.norm.bias, backbone.encoder_block2.0.norm.weight, backbone.encoder_block2.0.conv.weight, backbone.encoder_block1.1.attn.to_out.0.bias, backbone.encoder_block1.1.attn.to_out.0.weight, backbone.encoder_block1.1.attn.to_v.weight, backbone.encoder_block1.1.attn.to_k.weight, backbone.encoder_block1.1.attn.to_q.weight, backbone.encoder_block1.1.norm.bias, backbone.encoder_block1.1.norm.weight, backbone.encoder_block1.1.resnet.conv2.bias, backbone.encoder_block1.1.resnet.conv2.weight, backbone.encoder_block1.1.resnet.norm2.bias, backbone.encoder_block1.1.resnet.norm2.weight, backbone.encoder_block1.1.resnet.conv1.bias, backbone.encoder_block1.1.resnet.conv1.weight, backbone.encoder_block1.1.resnet.norm1.bias, backbone.encoder_block1.1.resnet.norm1.weight, backbone.encoder_block1.0.attn.to_out.0.bias, backbone.encoder_block1.0.attn.to_out.0.weight, backbone.encoder_block1.0.attn.to_v.weight, backbone.encoder_block1.0.attn.to_k.weight, backbone.encoder_block1.0.attn.to_q.weight, backbone.encoder_block1.0.norm.bias, backbone.encoder_block1.0.norm.weight, backbone.encoder_block1.0.resnet.conv2.bias, backbone.encoder_block1.0.resnet.conv2.weight, backbone.encoder_block1.0.resnet.norm2.bias, backbone.encoder_block1.0.resnet.norm2.weight, backbone.encoder_block1.0.resnet.conv1.bias, backbone.encoder_block1.0.resnet.conv1.weight, backbone.encoder_block1.0.resnet.norm1.bias, backbone.encoder_block1.0.resnet.norm1.weight, backbone.conv1x1.bias, backbone.conv1x1.weight [rank0]: Parameter indices which did not receive grad for rank 0: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67
05-14
用户已经知道可以通过设置find_unused_parameters=True来暂时解决，但可能希望更根本的解决方法，或者想了解如何避免这种情况。 首先，我需要回忆DDP的工作原理。DDP在每次反向传播后会同步所有参与计算的参数的...
--------------------------------------------------------------------------- RuntimeError Traceback (most recent call last) Cell In[6], line 18 15 labels = input_ids.clone() 17 # 训练模式 ---> 18 output = model(input_ids=input_ids, labels=labels) 19 print("Loss:", output["loss"].item()) 20 print("Logits shape:", output["logits"].shape) File f:\Programmer\python\MyAI\.venv\Lib\site-packages\torch\nn\modules\module.py:1775, in Module._wrapped_call_impl(self, *args, **kwargs) 1773 return self._compiled_call_impl(*args, **kwargs) # type: ignore[misc] 1774 else: -> 1775 return self._call_impl(*args, **kwargs) File f:\Programmer\python\MyAI\.venv\Lib\site-packages\torch\nn\modules\module.py:1786, in Module._call_impl(self, *args, **kwargs) 1781 # If we don't have any hooks, we want to skip the rest of the logic in 1782 # this function, and just call forward. 1783 if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks 1784 or _global_backward_pre_hooks or _global_backward_hooks 1785 or _global_forward_hooks or _global_forward_pre_hooks): -> 1786 return forward_call(*args, **kwargs) 1788 result = None 1789 called_always_called_hooks = set() Cell In[5], line 60, in YiziLM.forward(self, input_ids, labels) 58 hidden_states = inputs_embeds 59 for layer in self.layers: ---> 60 hidden_states = layer(hidden_states, freqs_cis) 62 hidden_states = self.norm(hidden_states) 63 logits = self.lm_head(hidden_states) File f:\Programmer\python\MyAI\.venv\Lib\site-packages\torch\nn\modules\module.py:1775, in Module._wrapped_call_impl(self, *args, **kwargs) 1773 return self._compiled_call_impl(*args, **kwargs) # type: ignore[misc] 1774 else: -> 1775 return self._call_impl(*args, **kwargs) File f:\Programmer\python\MyAI\.venv\Lib\site-packages\torch\nn\modules\module.py:1786, in Module._call_impl(self, *args, **kwargs) 1781 # If we don't have any hooks, we want to skip the rest of the logic in 1782 # this function, and just call forward. 1783 if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks 1784 or _global_backward_pre_hooks or _global_backward_hooks 1785 or _global_forward_hooks or _global_forward_pre_hooks): -> 1786 return forward_call(*args, **kwargs) 1788 result = None 1789 called_always_called_hooks = set() Cell In[4], line 15, in YiziBlock.forward(self, x, freqs_cis) 13 def forward(self, x: torch.Tensor, freqs_cis: torch.Tensor) -> torch.Tensor: 14 # 注意力残差连接 ---> 15 x = x + self.attn(self.norm1(x), freqs_cis) 16 # MLP 残差连接 17 x = x + self.mlp(self.norm2(x)) File f:\Programmer\python\MyAI\.venv\Lib\site-packages\torch\nn\modules\module.py:1775, in Module._wrapped_call_impl(self, *args, **kwargs) 1773 return self._compiled_call_impl(*args, **kwargs) # type: ignore[misc] 1774 else: -> 1775 return self._call_impl(*args, **kwargs) File f:\Programmer\python\MyAI\.venv\Lib\site-packages\torch\nn\modules\module.py:1786, in Module._call_impl(self, *args, **kwargs) 1781 # If we don't have any hooks, we want to skip the rest of the logic in 1782 # this function, and just call forward. 1783 if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks 1784 or _global_backward_pre_hooks or _global_backward_hooks 1785 or _global_forward_hooks or _global_forward_pre_hooks): -> 1786 return forward_call(*args, **kwargs) 1788 result = None 1789 called_always_called_hooks = set() Cell In[3], line 23, in YiziAttention.forward(self, x, freqs_cis) 20 value_states = self.v_proj(x).view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2) 22 # 应用 RoPE ---> 23 query_states, key_states = apply_rotary_pos_emb(query_states, key_states, freqs_cis) 25 # Scaled Dot-Product Attention 26 attn_weights = torch.matmul(query_states, key_states.transpose(2, 3)) * self.scale Cell In[2], line 51, in apply_rotary_pos_emb(q, k, freqs_cis) 47 freqs_cis = freqs_cis[:seq_len].view(1, 1, seq_len, -1, 2) 49 # 复数乘法: (a + bi)(c + di) = (ac - bd) + (ad + bc)i 50 # 这里使用实数模拟复数运算 ---> 51 q_out_real = q_[..., 0] * freqs_cis[..., 0] - q_[..., 1] * freqs_cis[..., 1] 52 q_out_imag = q_[..., 0] * freqs_cis[..., 1] + q_[..., 1] * freqs_cis[..., 0] 53 q_out = torch.stack([q_out_real, q_out_imag], dim=-1).flatten(-2) RuntimeError: The size of tensor a (32) must match the size of tensor b (16) at non-singleton dimension 3
最新发布
12-01
self.norm = nn.LayerNorm(config.hidden_size) self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False) self.lm_head.weight = self.embed_tokens.weight ``` --- ### ✅ Step 3: 修复...
x=self.sublayer[0](x,lambda x:self.self_attn(x,x,x,mask)) return x+self.dropout(sublayer(self.norm(x))) el_result=el(x,mask) __init__() takes from 3 to 4 positional arguments but 5 were given
09-15
但这里`sublayer(self.norm(x))`只传递了一个参数（self.norm(x)），而`SublayerConnection`的`__call__`需要两个参数（self, x, sublayer_fn）。所以这里调用方式错误。 但是，这个错误可能不是当前报告的错误，...
#下面代码运行时报错： "C:\Program Files\Python312\python.exe" C:\Users\Administrator\py2025\GPS对抗程序A.py Traceback (most recent call last): File "C:\Users\Administrator\py2025\GPS对抗程序A.py", line 858, in <module> window = GPSJammingSimulator() ^^^^^^^^^^^^^^^^^^^^^ File "C:\Users\Administrator\py2025\GPS对抗程序A.py", line 221, in __init__ self.update_simulation() File "C:\Users\Administrator\py2025\GPS对抗程序A.py", line 592, in update_simulation t, signal = self.gps_system.generate_gps_signal() ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\Users\Administrator\py2025\GPS对抗程序A.py", line 65, in generate_gps_signal return t, signal ^^^^^^ NameError: name 'signal' is not defined. Did you forget to import 'signal'? 进程已结束，退出代码为 1 ========================================================================================== import sys import numpy as np from PyQt5.QtWidgets import (QApplication, QMainWindow, QWidget, QVBoxLayout, QHBoxLayout, QGridLayout, QGroupBox, QLabel, QSlider, QComboBox, QPushButton, QDoubleSpinBox, QCheckBox, QFrame, QDoubleSpinBox, QCheckBox, QFrame,QSpinBox) from PyQt5.QtCore import Qt, QTimer from PyQt5.QtGui import QFont, QPalette, QColor import matplotlib.pyplot as plt from matplotlib.backends.backend_qt5agg import FigureCanvasQTAgg as FigureCanvas from matplotlib.figure import Figure # 添加中文字体支持 plt.rcParams['font.sans-serif'] = ['SimHei', 'Microsoft YaHei', 'SimSun', 'Arial Unicode MS'] # 设置中文字体 plt.rcParams['axes.unicode_minus'] = False # 解决负号显示问题 class GPSSystem: def __init__(self): # GPS信号参数（L1频段） self.satellite_id = 1 # 卫星编号 self.carrier_freq = 1575.42 # L1频段中心频率(MHz) self.signal_power = -160 # dBW self.code_rate = 1.023 # C/A码速率(Mcps) self.data_rate = 50 # 导航数据速率(bps) self.modulation_type = "BPSK" # 调制方式 # 环境参数 self.iono_delay = 5 # 电离层延迟(ns) self.tropo_delay = 1.5 # 对流层延迟(ns) self.multipath = 0.5 # 多径效应幅度 # 干扰参数 self.jamming_type = "None" # None, Narrowband, Pulse, Spoofing, Swept self.jamming_power = -100 # dBW self.jamming_freq = 1575.42 # 干扰频率(MHz) self.spoofing_delay = 0.1 # 欺骗干扰延迟(ms) self.spoofing_count = 1 # 虚假卫星数量 # 系统参数 self.sample_rate = 100 # 采样率(MS/s) self.duration = 1 # 信号持续时间(ms) self.noise_power = -200 # 热噪声功率(dBW) # 卫星信息 self.satellite_pos = np.array([12000, 20000, 5000]) # 卫星位置(km) self.user_pos = np.array([0, 0, 0]) # 用户位置(km) def generate_gps_signal(self): t = np.linspace(0, self.duration, int(self.sample_rate * self.duration * 1e3)) # 生成导航数据（修复除零错误） num_bits = max(1, int(self.data_rate * self.duration * 1e-3)) # 确保至少1个比特 data_bits = np.random.choice([-1, 1], size=num_bits) # 计算重复次数（避免除零） repeat_count = len(t) // num_bits if num_bits > 0 else len(t) data_signal = np.repeat(data_bits, repeat_count) # 截断到与t相同长度 if len(data_signal) > len(t): data_signal = data_signal[:len(t)] # 其余代码保持不变... return t, signal def generate_interference(self, t): """生成干扰信号""" if self.jamming_type == "None": return np.zeros_like(t) elif self.jamming_type == "Narrowband": # 窄带干扰 phase = 2 * np.pi * self.jamming_freq * t * 1e6 * 1e-3 return 10 ** (self.jamming_power / 20) * np.sin(phase) elif self.jamming_type == "Pulse": # 脉冲干扰 pulse_width = 0.1 # 脉冲宽度(ms) pulse_start = np.random.uniform(0, self.duration - pulse_width) pulse = np.zeros_like(t) pulse[(t >= pulse_start) & (t <= pulse_start + pulse_width)] = 1 phase = 2 * np.pi * self.jamming_freq * t * 1e6 * 1e-3 return 10 ** (self.jamming_power / 20) * pulse * np.sin(phase) elif self.jamming_type == "Swept": # 扫频干扰 sweep_rate = 10 # 扫频速率(MHz/ms) freq = self.jamming_freq + sweep_rate * t phase = 2 * np.pi * freq * t * 1e6 * 1e-3 return 10 ** (self.jamming_power / 20) * np.sin(phase) elif self.jamming_type == "Spoofing": # 欺骗干扰 spoofing_signal = np.zeros_like(t) for i in range(self.spoofing_count): delay = self.spoofing_delay * (i + 1) phase = 2 * np.pi * self.carrier_freq * (t - delay) * 1e6 * 1e-3 spoofing_signal += 10 ** ((self.jamming_power - 3) / 20) * np.sin(phase) return spoofing_signal return np.zeros_like(t) def add_noise(self, signal): """添加热噪声""" noise_power_linear = 10 ** (self.noise_power / 10) noise = np.random.normal(0, np.sqrt(noise_power_linear / 2), len(signal)) * 1j noise += np.random.normal(0, np.sqrt(noise_power_linear / 2), len(signal)) return signal + noise.real def add_multipath(self, signal): """添加多径效应""" if self.multipath > 0: delay_samples = int(0.1 * self.sample_rate * 1e3 * self.duration / 1000) multipath_signal = np.roll(signal, delay_samples) multipath_signal[:delay_samples] = 0 return signal + self.multipath * multipath_signal return signal def calculate_pseudorange(self, signal): """计算伪距""" # 简化的伪距计算 noise_signal = self.add_noise(signal) interference_signal = self.generate_interference(t) combined_signal = noise_signal + interference_signal # 简单的信号检测 correlation = np.abs(np.correlate(combined_signal, signal[:100], mode='valid')) peak_index = np.argmax(correlation) # 计算伪距 pseudorange = (peak_index / (self.sample_rate * 1e6)) * 3e8 / 1e3 # km pseudorange += self.iono_delay * 0.3 # 电离层修正 pseudorange += self.tropo_delay * 0.1 # 对流层修正 return pseudorange def calculate_position(self): """计算位置""" # 简单的多卫星位置计算 true_position = self.user_pos estimated_position = np.zeros(3) # 模拟多个卫星观测 for _ in range(4): # 至少需要4颗卫星 self.satellite_pos = np.random.uniform(10000, 20000, 3) pseudorange = self.calculate_pseudorange(self.generate_gps_signal()[1]) estimated_position += (self.satellite_pos - (pseudorange / np.linalg.norm(self.satellite_pos)) * (self.satellite_pos - true_position)) return estimated_position / 4 def calculate_dop(self): """计算几何精度因子""" # 简化的DOP计算 satellite_positions = np.random.uniform(10000, 20000, (8, 3)) geometry_matrix = np.zeros((8, 4)) for i, sat_pos in enumerate(satellite_positions): sat_dist = np.linalg.norm(sat_pos) geometry_matrix[i, :3] = sat_pos / sat_dist geometry_matrix[i, 3] = 1 try: q_matrix = np.linalg.inv(geometry_matrix.T @ geometry_matrix) pdop = np.sqrt(q_matrix[0, 0] + q_matrix[1, 1] + q_matrix[2, 2]) hdop = np.sqrt(q_matrix[0, 0] + q_matrix[1, 1]) vdop = np.sqrt(q_matrix[2, 2]) except np.linalg.LinAlgError: pdop, hdop, vdop = 99.9, 99.9, 99.9 return pdop, hdop, vdop class GPSJammingSimulator(QMainWindow): def __init__(self): super().__init__() self.setWindowTitle("GPS对抗仿真系统") self.setGeometry(100, 50, 1400, 900) # 初始化GPS系统 self.gps_system = GPSSystem() # 设置深色主题 self.set_dark_theme() # 创建主控件 self.main_widget = QWidget() self.setCentralWidget(self.main_widget) # 主布局 main_layout = QVBoxLayout(self.main_widget) main_layout.setSpacing(15) main_layout.setContentsMargins(20, 20, 20, 20) # 标题 title_label = QLabel("GPS对抗仿真系统") title_label.setFont(QFont("Arial", 24, QFont.Bold)) title_label.setStyleSheet("color: #1E90FF;") title_label.setAlignment(Qt.AlignCenter) main_layout.addWidget(title_label) # 创建控制面板 control_panel = self.create_control_panel() main_layout.addWidget(control_panel) # 创建信号显示区域 signal_panel = self.create_signal_panel() main_layout.addWidget(signal_panel, 1) # 创建位置显示区域 position_panel = self.create_position_panel() main_layout.addWidget(position_panel) # 创建状态显示区域 status_panel = self.create_status_panel() main_layout.addWidget(status_panel) # 初始化数据 self.update_simulation() # 设置定时器用于实时更新 self.timer = QTimer(self) self.timer.timeout.connect(self.update_simulation) self.timer.start(100) # 每100ms更新一次 def set_dark_theme(self): """设置深色主题""" dark_palette = QPalette() dark_palette.setColor(QPalette.Window, QColor(25, 35, 45)) dark_palette.setColor(QPalette.WindowText, Qt.white) dark_palette.setColor(QPalette.Base, QColor(35, 45, 55)) dark_palette.setColor(QPalette.AlternateBase, QColor(45, 55, 65)) dark_palette.setColor(QPalette.ToolTipBase, Qt.white) dark_palette.setColor(QPalette.ToolTipText, Qt.white) dark_palette.setColor(QPalette.Text, Qt.white) dark_palette.setColor(QPalette.Button, QColor(50, 65, 80)) dark_palette.setColor(QPalette.ButtonText, Qt.white) dark_palette.setColor(QPalette.BrightText, Qt.red) dark_palette.setColor(QPalette.Highlight, QColor(30, 144, 255)) dark_palette.setColor(QPalette.HighlightedText, Qt.black) self.setPalette(dark_palette) self.setStyleSheet(""" QGroupBox { font-size: 14px; font-weight: bold; border: 2px solid #1E90FF; border-radius: 8px; margin-top: 1ex; color: #1E90FF; } QGroupBox::title { subcontrol-origin: margin; subcontrol-position: top center; padding: 0 5px; } QSlider::groove:horizontal { border: 1px solid #4A708B; height: 8px; background: #2F4F4F; margin: 2px 0; border-radius: 4px; } QSlider::handle:horizontal { background: #1E90FF; border: 1px solid #4A708B; width: 18px; margin: -4px 0; border-radius: 9px; } QPushButton { background-color: #1E90FF; color: white; border: none; border-radius: 4px; padding: 5px 10px; font-weight: bold; } QPushButton:hover { background-color: #00BFFF; } QLabel { color: white; } """) def create_control_panel(self): """创建控制面板""" panel = QGroupBox("GPS对抗控制面板") layout = QGridLayout() panel.setLayout(layout) # 卫星参数控制 satellite_label = QLabel("卫星参数:") satellite_label.setFont(QFont("Arial", 12, QFont.Bold)) layout.addWidget(satellite_label, 0, 0) # 卫星编号 sat_id_label = QLabel("卫星编号:") self.sat_id_spin = QSpinBox() self.sat_id_spin.setRange(1, 32) self.sat_id_spin.setValue(self.gps_system.satellite_id) self.sat_id_spin.valueChanged.connect(self.update_sat_id) layout.addWidget(sat_id_label, 1, 0) layout.addWidget(self.sat_id_spin, 1, 1) # 载波频率 freq_label = QLabel("载波频率 (MHz):") self.freq_spin = QDoubleSpinBox() self.freq_spin.setRange(1570, 1580) self.freq_spin.setValue(self.gps_system.carrier_freq) self.freq_spin.setSingleStep(0.01) self.freq_spin.valueChanged.connect(self.update_carrier_freq) layout.addWidget(freq_label, 2, 0) layout.addWidget(self.freq_spin, 2, 1) # 信号功率 power_label = QLabel("信号功率 (dBW):") self.power_slider = QSlider(Qt.Horizontal) self.power_slider.setRange(-170, -150) self.power_slider.setValue(int(self.gps_system.signal_power)) self.power_slider.valueChanged.connect(self.update_signal_power) layout.addWidget(power_label, 3, 0) layout.addWidget(self.power_slider, 3, 1) # 干扰参数控制 jam_label = QLabel("干扰参数:") jam_label.setFont(QFont("Arial", 12, QFont.Bold)) layout.addWidget(jam_label, 0, 2) # 干扰类型 jam_type_label = QLabel("干扰类型:") self.jam_type_combo = QComboBox() self.jam_type_combo.addItems(["None", "Narrowband", "Pulse", "Spoofing", "Swept"]) self.jam_type_combo.setCurrentText(self.gps_system.jamming_type) self.jam_type_combo.currentTextChanged.connect(self.update_jam_type) layout.addWidget(jam_type_label, 1, 2) layout.addWidget(self.jam_type_combo, 1, 3) # 干扰功率 jam_power_label = QLabel("干扰功率 (dBW):") self.jam_power_slider = QSlider(Qt.Horizontal) self.jam_power_slider.setRange(-150, -50) self.jam_power_slider.setValue(int(self.gps_system.jamming_power)) self.jam_power_slider.valueChanged.connect(self.update_jam_power) layout.addWidget(jam_power_label, 2, 2) layout.addWidget(self.jam_power_slider, 2, 3) # 干扰频率 jam_freq_label = QLabel("干扰频率 (MHz):") self.jam_freq_spin = QDoubleSpinBox() self.jam_freq_spin.setRange(1570, 1580) self.jam_freq_spin.setValue(self.gps_system.jamming_freq) self.jam_freq_spin.setSingleStep(0.01) self.jam_freq_spin.valueChanged.connect(self.update_jam_freq) layout.addWidget(jam_freq_label, 3, 2) layout.addWidget(self.jam_freq_spin, 3, 3) # 欺骗干扰参数 spoofing_label = QLabel("欺骗干扰参数:") spoofing_label.setFont(QFont("Arial", 12, QFont.Bold)) layout.addWidget(spoofing_label, 0, 4) # 欺骗延迟 spoofing_delay_label = QLabel("欺骗延迟 (ms):") self.spoofing_delay_spin = QDoubleSpinBox() self.spoofing_delay_spin.setRange(0.01, 1.0) self.spoofing_delay_spin.setValue(self.gps_system.spoofing_delay) self.spoofing_delay_spin.setSingleStep(0.01) self.spoofing_delay_spin.valueChanged.connect(self.update_spoofing_delay) layout.addWidget(spoofing_delay_label, 1, 4) layout.addWidget(self.spoofing_delay_spin, 1, 5) # 虚假卫星数量 spoofing_count_label = QLabel("虚假卫星数量:") self.spoofing_count_spin = QSpinBox() self.spoofing_count_spin.setRange(1, 5) self.spoofing_count_spin.setValue(self.gps_system.spoofing_count) self.spoofing_count_spin.valueChanged.connect(self.update_spoofing_count) layout.addWidget(spoofing_count_label, 2, 4) layout.addWidget(self.spoofing_count_spin, 2, 5) # 环境参数控制 env_label = QLabel("环境参数:") env_label.setFont(QFont("Arial", 12, QFont.Bold)) layout.addWidget(env_label, 5, 0) # 电离层延迟 iono_label = QLabel("电离层延迟 (ns):") self.iono_slider = QSlider(Qt.Horizontal) self.iono_slider.setRange(0, 20) self.iono_slider.setValue(int(self.gps_system.iono_delay)) self.iono_slider.valueChanged.connect(self.update_iono_delay) layout.addWidget(iono_label, 6, 0) layout.addWidget(self.iono_slider, 6, 1) # 多径效应 multipath_label = QLabel("多径效应幅度:") self.multipath_slider = QSlider(Qt.Horizontal) self.multipath_slider.setRange(0, 100) self.multipath_slider.setValue(int(self.gps_system.multipath * 100)) self.multipath_slider.valueChanged.connect(self.update_multipath) layout.addWidget(multipath_label, 7, 0) layout.addWidget(self.multipath_slider, 7, 1) # 系统参数 sys_label = QLabel("系统参数:") sys_label.setFont(QFont("Arial", 12, QFont.Bold)) layout.addWidget(sys_label, 5, 2) # 采样率 sample_rate_label = QLabel("采样率 (MS/s):") self.sample_rate_spin = QDoubleSpinBox() self.sample_rate_spin.setRange(50, 200) self.sample_rate_spin.setValue(self.gps_system.sample_rate) self.sample_rate_spin.setSingleStep(10) self.sample_rate_spin.valueChanged.connect(self.update_sample_rate) layout.addWidget(sample_rate_label, 6, 2) layout.addWidget(self.sample_rate_spin, 6, 3) # 噪声功率 noise_label = QLabel("热噪声功率 (dBW):") self.noise_slider = QSlider(Qt.Horizontal) self.noise_slider.setRange(-210, -150) self.noise_slider.setValue(int(self.gps_system.noise_power)) self.noise_slider.valueChanged.connect(self.update_noise_power) layout.addWidget(noise_label, 7, 2) layout.addWidget(self.noise_slider, 7, 3) # 控制按钮 self.reset_btn = QPushButton("重置参数") self.reset_btn.clicked.connect(self.reset_params) layout.addWidget(self.reset_btn, 8, 0, 1, 2) self.auto_jam_btn = QPushButton("自动干扰") self.auto_jam_btn.clicked.connect(self.toggle_auto_jam) layout.addWidget(self.auto_jam_btn, 8, 2, 1, 2) self.optimize_btn = QPushButton("优化定位") self.optimize_btn.clicked.connect(self.optimize_positioning) layout.addWidget(self.optimize_btn, 8, 4, 1, 2) self.auto_jam_active = False return panel def create_signal_panel(self): """创建信号显示面板""" panel = QGroupBox("GPS信号与干扰信号") layout = QHBoxLayout() panel.setLayout(layout) # 原始信号波形图 self.signal_fig = Figure(figsize=(10, 4), dpi=100) self.signal_canvas = FigureCanvas(self.signal_fig) self.signal_ax = self.signal_fig.add_subplot(111) self.signal_ax.set_title("原始GPS信号") self.signal_ax.set_xlabel("时间 (ms)") self.signal_ax.set_ylabel("幅度") self.signal_ax.grid(True, linestyle='--', alpha=0.6) layout.addWidget(self.signal_canvas) # 干扰信号波形图 self.jamming_fig = Figure(figsize=(10, 4), dpi=100) self.jamming_canvas = FigureCanvas(self.jamming_fig) self.jamming_ax = self.jamming_fig.add_subplot(111) self.jamming_ax.set_title("干扰信号") self.jamming_ax.set_xlabel("时间 (ms)") self.jamming_ax.set_ylabel("幅度") self.jamming_ax.grid(True, linestyle='--', alpha=0.6) layout.addWidget(self.jamming_canvas) # 合成信号波形图 self.combined_fig = Figure(figsize=(10, 4), dpi=100) self.combined_canvas = FigureCanvas(self.combined_fig) self.combined_ax = self.combined_fig.add_subplot(111) self.combined_ax.set_title("合成信号 (GPS + 干扰)") self.combined_ax.set_xlabel("时间 (ms)") self.combined_ax.set_ylabel("幅度") self.combined_ax.grid(True, linestyle='--', alpha=0.6) layout.addWidget(self.combined_canvas) return panel def create_position_panel(self): """创建位置显示面板""" panel = QGroupBox("定位信息") layout = QHBoxLayout() panel.setLayout(layout) # 三维位置图 self.position_fig = Figure(figsize=(8, 6), dpi=100) self.position_canvas = FigureCanvas(self.position_fig) self.position_ax = self.position_fig.add_subplot(111, projection='3d') self.position_ax.set_title("卫星与用户位置") self.position_ax.set_xlabel("X (km)") self.position_ax.set_ylabel("Y (km)") self.position_ax.set_zlabel("Z (km)") self.position_ax.grid(True, linestyle='--', alpha=0.6) layout.addWidget(self.position_canvas) # 定位误差图 self.error_fig = Figure(figsize=(6, 6), dpi=100) self.error_canvas = FigureCanvas(self.error_fig) self.error_ax = self.error_fig.add_subplot(111) self.error_ax.set_title("定位误差分布") self.error_ax.set_xlabel("X误差 (m)") self.error_ax.set_ylabel("Y误差 (m)") self.error_ax.grid(True, linestyle='--', alpha=0.6) layout.addWidget(self.error_canvas) return panel def create_status_panel(self): """创建状态显示面板""" panel = QFrame() panel.setFrameShape(QFrame.StyledPanel) layout = QHBoxLayout() panel.setLayout(layout) # 定位状态 position_status = QGroupBox("定位状态") pos_layout = QVBoxLayout() position_status.setLayout(pos_layout) self.position_label = QLabel("定位模式: 三维定位") self.position_label.setFont(QFont("Arial", 14)) self.position_label.setStyleSheet("color: #00FF00;") pos_layout.addWidget(self.position_label) self.pdop_label = QLabel("PDOP: 1.0") self.pdop_label.setFont(QFont("Arial", 12)) pos_layout.addWidget(self.pdop_label) self.hdop_label = QLabel("HDOP: 0.8") self.hdop_label.setFont(QFont("Arial", 12)) pos_layout.addWidget(self.hdop_label) self.vdop_label = QLabel("VDOP: 0.6") self.vdop_label.setFont(QFont("Arial", 12)) pos_layout.addWidget(self.vdop_label) layout.addWidget(position_status) # 干扰状态 jam_status = QGroupBox("干扰状态") jam_layout = QVBoxLayout() jam_status.setLayout(jam_layout) self.jam_status_label = QLabel("干扰强度: 弱") self.jam_status_label.setFont(QFont("Arial", 14)) self.jam_status_label.setStyleSheet("color: #FFA500;") jam_layout.addWidget(self.jam_status_label) self.jam_power_label = QLabel("干扰功率: -100 dBW") self.jam_power_label.setFont(QFont("Arial", 12)) jam_layout.addWidget(self.jam_power_label) self.jam_effect_label = QLabel("干扰效果: 无") self.jam_effect_label.setFont(QFont("Arial", 12)) jam_layout.addWidget(self.jam_effect_label) layout.addWidget(jam_status) # 系统状态 sys_status = QGroupBox("系统状态") sys_layout = QVBoxLayout() sys_status.setLayout(sys_layout) self.sys_status_label = QLabel("系统运行: 正常") self.sys_status_label.setFont(QFont("Arial", 14)) self.sys_status_label.setStyleSheet("color: #00FF00;") sys_layout.addWidget(self.sys_status_label) self.sample_rate_status = QLabel("采样率: 100 MS/s") self.sample_rate_status.setFont(QFont("Arial", 12)) sys_layout.addWidget(self.sample_rate_status) self.freq_range_status = QLabel("频率范围: 1570-1580 MHz") self.freq_range_status.setFont(QFont("Arial", 12)) sys_layout.addWidget(self.freq_range_status) layout.addWidget(sys_status) return panel def update_simulation(self): """更新仿真数据并刷新界面""" # 生成信号数据 t, signal = self.gps_system.generate_gps_signal() jamming_signal = self.gps_system.generate_interference(t) combined_signal = signal + jamming_signal # 计算定位信息 estimated_pos = self.gps_system.calculate_position() pdop, hdop, vdop = self.gps_system.calculate_dop() # 计算干扰强度 jam_power = 10 ** (self.gps_system.jamming_power / 10) # 更新波形图 self.update_waveform_plot(t, signal, jamming_signal, combined_signal) # 更新位置图 self.update_position_plot(estimated_pos) # 更新状态显示 self.update_status(jam_power, pdop, hdop, vdop) # 自动干扰模式 if self.auto_jam_active: self.auto_jamming() def update_waveform_plot(self, t, signal, jamming_signal, combined_signal): """更新波形图""" # 原始信号 self.signal_ax.clear() self.signal_ax.plot(t, signal, 'b-', linewidth=1.5) self.signal_ax.set_title("原始GPS信号") self.signal_ax.set_xlabel("时间 (ms)") self.signal_ax.set_ylabel("幅度") self.signal_ax.grid(True, linestyle='--', alpha=0.6) self.signal_ax.set_xlim(0, self.gps_system.duration) self.signal_canvas.draw() # 干扰信号 self.jamming_ax.clear() self.jamming_ax.plot(t, jamming_signal, 'r-', linewidth=1.5) self.jamming_ax.set_title("干扰信号") self.jamming_ax.set_xlabel("时间 (ms)") self.jamming_ax.set_ylabel("幅度") self.jamming_ax.grid(True, linestyle='--', alpha=0.6) self.jamming_ax.set_xlim(0, self.gps_system.duration) self.jamming_canvas.draw() # 合成信号 self.combined_ax.clear() self.combined_ax.plot(t, combined_signal, 'g-', linewidth=1.5) self.combined_ax.set_title("合成信号 (GPS + 干扰)") self.combined_ax.set_xlabel("时间 (ms)") self.combined_ax.set_ylabel("幅度") self.combined_ax.grid(True, linestyle='--', alpha=0.6) self.combined_ax.set_xlim(0, self.gps_system.duration) self.combined_canvas.draw() def update_position_plot(self, estimated_pos): """更新位置图""" # 清除旧的图形 self.position_ax.clear() # 设置标题和标签 self.position_ax.set_title("卫星与用户位置") self.position_ax.set_xlabel("X (km)") self.position_ax.set_ylabel("Y (km)") self.position_ax.set_zlabel("Z (km)") self.position_ax.grid(True, linestyle='--', alpha=0.6) # 设置坐标轴范围 self.position_ax.set_xlim(0, 25000) self.position_ax.set_ylim(0, 25000) self.position_ax.set_zlim(0, 25000) # 绘制卫星位置 satellite_pos = self.gps_system.satellite_pos self.position_ax.scatter(satellite_pos[0], satellite_pos[1], satellite_pos[2], c='red', s=100, marker='^', label='卫星') # 绘制用户真实位置 user_pos = self.gps_system.user_pos self.position_ax.scatter(user_pos[0], user_pos[1], user_pos[2], c='blue', s=100, marker='o', label='真实位置') # 绘制估计位置 self.position_ax.scatter(estimated_pos[0], estimated_pos[1], estimated_pos[2], c='green', s=100, marker='x', label='估计位置') # 添加图例 self.position_ax.legend() # 更新画布 self.position_canvas.draw() # 更新误差图 error = estimated_pos - user_pos self.error_ax.clear() self.error_ax.plot(error[0] * 1000, error[1] * 1000, 'ro', markersize=10) self.error_ax.set_title("定位误差分布") self.error_ax.set_xlabel("X误差 (m)") self.error_ax.set_ylabel("Y误差 (m)") self.error_ax.axis('equal') self.error_ax.grid(True, linestyle='--', alpha=0.6) self.error_ax.set_xlim(-100, 100) self.error_ax.set_ylim(-100, 100) self.error_canvas.draw() def update_status(self, jam_power, pdop, hdop, vdop): """更新状态显示""" # 更新DOP值 self.pdop_label.setText(f"PDOP: {pdop:.2f}") self.hdop_label.setText(f"HDOP: {hdop:.2f}") self.vdop_label.setText(f"VDOP: {vdop:.2f}") # 定位精度评估 if pdop < 2: self.position_label.setText("定位模式: 精密定位") self.position_label.setStyleSheet("color: #00FF00;") elif pdop < 5: self.position_label.setText("定位模式: 良好定位") self.position_label.setStyleSheet("color: #7CFC00;") elif pdop < 10: self.position_label.setText("定位模式: 一般定位") self.position_label.setStyleSheet("color: #FFD700;") else: self.position_label.setText("定位模式: 低精度定位") self.position_label.setStyleSheet("color: #FF4500;") # 干扰状态 self.jam_power_label.setText(f"干扰功率: {self.gps_system.jamming_power} dBW") if jam_power < 1e-12: # -120 dBW self.jam_status_label.setText("干扰强度: 无") self.jam_status_label.setStyleSheet("color: #00FF00;") elif jam_power < 1e-10: # -100 dBW self.jam_status_label.setText("干扰强度: 弱") self.jam_status_label.setStyleSheet("color: #7CFC00;") elif jam_power < 1e-8: # -80 dBW self.jam_status_label.setText("干扰强度: 中") self.jam_status_label.setStyleSheet("color: #FFA500;") else: self.jam_status_label.setText("干扰强度: 强") self.jam_status_label.setStyleSheet("color: #FF4500;") # 干扰效果评估 if pdop > 10: self.jam_effect_label.setText("干扰效果: 严重") elif pdop > 5: self.jam_effect_label.setText("干扰效果: 显著") elif pdop > 2: self.jam_effect_label.setText("干扰效果: 轻微") else: self.jam_effect_label.setText("干扰效果: 无") # 系统状态 self.sample_rate_status.setText(f"采样率: {self.gps_system.sample_rate} MS/s") def auto_jamming(self): """自动干扰模式""" # 随机改变干扰参数 if np.random.rand() < 0.1: self.jam_type_combo.setCurrentIndex(np.random.randint(1, 5)) if np.random.rand() < 0.2: new_freq = max(1570, min(1580, self.gps_system.jamming_freq + np.random.uniform(-0.5, 0.5))) self.jam_freq_spin.setValue(new_freq) if np.random.rand() < 0.3: new_power = max(-150, min(-50, self.gps_system.jamming_power + np.random.randint(-5, 5))) self.jam_power_slider.setValue(int(new_power)) if self.gps_system.jamming_type == "Spoofing" and np.random.rand() < 0.2: new_delay = max(0.01, min(1.0, self.gps_system.spoofing_delay + np.random.uniform(-0.05, 0.05))) self.spoofing_delay_spin.setValue(new_delay) if self.gps_system.jamming_type == "Spoofing" and np.random.rand() < 0.1: new_count = max(1, min(5, self.gps_system.spoofing_count + np.random.randint(-1, 2))) self.spoofing_count_spin.setValue(new_count) def toggle_auto_jam(self): """切换自动干扰模式""" self.auto_jam_active = not self.auto_jam_active if self.auto_jam_active: self.auto_jam_btn.setText("停止干扰") self.auto_jam_btn.setStyleSheet("background-color: #FF4500;") else: self.auto_jam_btn.setText("自动干扰") self.auto_jam_btn.setStyleSheet("") def optimize_positioning(self): """优化定位参数""" # 简单优化策略：避开干扰频率 if self.gps_system.jamming_type != "None" and self.gps_system.jamming_freq > 0: # 尝试调整卫星编号以避开干扰 new_id = (self.gps_system.satellite_id % 32) + 1 self.sat_id_spin.setValue(new_id) self.position_label.setText("定位模式: 卫星切换中") def reset_params(self): """重置参数到默认值""" self.gps_system = GPSSystem() # 重置卫星参数 self.sat_id_spin.setValue(self.gps_system.satellite_id) self.freq_spin.setValue(self.gps_system.carrier_freq) self.power_slider.setValue(int(self.gps_system.signal_power)) # 重置干扰参数 self.jam_type_combo.setCurrentText(self.gps_system.jamming_type) self.jam_power_slider.setValue(int(self.gps_system.jamming_power)) self.jam_freq_spin.setValue(self.gps_system.jamming_freq) self.spoofing_delay_spin.setValue(self.gps_system.spoofing_delay) self.spoofing_count_spin.setValue(self.gps_system.spoofing_count) # 重置环境参数 self.iono_slider.setValue(int(self.gps_system.iono_delay)) self.multipath_slider.setValue(int(self.gps_system.multipath * 100)) # 重置系统参数 self.sample_rate_spin.setValue(self.gps_system.sample_rate) self.noise_slider.setValue(int(self.gps_system.noise_power)) # 重置自动干扰状态 self.auto_jam_active = False self.auto_jam_btn.setText("自动干扰") self.auto_jam_btn.setStyleSheet("") # 参数更新函数 def update_sat_id(self, value): self.gps_system.satellite_id = value def update_carrier_freq(self, value): self.gps_system.carrier_freq = value def update_signal_power(self, value): self.gps_system.signal_power = value def update_jam_type(self, text): self.gps_system.jamming_type = text def update_jam_power(self, value): self.gps_system.jamming_power = value def update_jam_freq(self, value): self.gps_system.jamming_freq = value def update_spoofing_delay(self, value): self.gps_system.spoofing_delay = value def update_spoofing_count(self, value): self.gps_system.spoofing_count = value def update_iono_delay(self, value): self.gps_system.iono_delay = value def update_multipath(self, value): self.gps_system.multipath = value / 100.0 def update_sample_rate(self, value): self.gps_system.sample_rate = value def update_noise_power(self, value): self.gps_system.noise_power = value if __name__ == "__main__": app = QApplication(sys.argv) window = GPSJammingSimulator() window.show() sys.exit(app.exec_())
10-13
我们分析错误：在generate_gps_signal函数中，返回了变量signal，但该变量在函数中并没有被定义。 查看代码，在generate_gps_signal函数中，我们生成了t和data_signal，但是最后返回的却是t和signal。 因此，问题...
Python 类中的__call__()方法
qq_43739296的博客
01-14 390
Python call()方法 在类中实现__call__()方法，可以在实例化一个该类的对象后，直接调用该对象，像使用函数一样，这样实现的就是__call__()方法中的内容。 举例： 太长不看版 def main(args): image_file_list = get_image_file_list(args.image_dir) text_recognizer = TextRecognizer(args) valid_image_file_list = [] img_
attention机制_【CV中的attention机制】
weixin_39664431的博客
11-27 313
前言: 本文介绍了一个用于语义分割领域的attention模块scSE。scSE模块与之前介绍的BAM模块很类似，不过在这里scSE模块只在语义分割中进行应用和测试，对语义分割准确率带来的提升比较大。提出scSE模块论文的全称是：《Concurrent Spatial and Channel ‘Squeeze & Excitation’ in Fully Convolutiona...
python中向量长度_向量的长度和单位长度_线性代数_day5
weixin_29285605的博客
03-02 910
import mathclass Vector:EPSION = 1e-8 # 1/10^8，数字足够的小def __init__(self, my_list):self._values = my_list@classmethoddef zero(cls, dim):"""返回一个dim维的零向量"""return cls([0] * dim)def norm(self):"""返回向量的模""...
transformer与vit代码阅读
m0_63156697的博客
09-29 741
无
Transformer代码完全解读！
Datawhale
09-15 6813
↑↑↑关注后"星标"Datawhale每日干货&每月组队学习，不错过Datawhale干货作者：安晟&闫永强，Datawhale成员本篇正文部分约1...
torch.norm()函数的用法
热门推荐
黄笳倞的博客
05-30 3万+
目录 一、函数定义 二、代码示例 三、整体代码 一、函数定义 公式： 意思就是inputs的一共N维的话对这N个数据求p范数，当然这个还是太抽象了，接下来还是看具体的代码~ p指的是求p范数的p值，函数默认p=2，那么就是求2范数 ...
双边格网(bilateral grid)
flow_specter的博客
07-24 8375
双边滤波速度很慢 —> 加速，双边网格。 首先明确一点，双边网格本质上是一个数据结构。 以单通道灰度值为例，双边网格结合了图像二维的空间域信息以及一维的灰度信息，可认为其是一个3D的数组。 举一个简单的例子，假设现在你手上有一只用于滤波/平滑的笔刷，当你用这只笔刷在图像EEE上的某一个位置(x,y)(x,y)(x,y)处点击了一下，对应的，3D双边网格的(x,y,E(x,y))(x,y,E(x,y))(x,y,E(x,y))位置将出现一个点，这个点即对应你在2D影像EEE上点击的那个点。随着这只笔刷
搞懂Transformer结构，看这篇PyTorch实现就够了！
qq_42370150的博客
03-07 2096
来源 | 哈工大SCIR "Attention is All You Need"[1] 一文中提出的Transformer网络结构最近引起了很多人的关注。Transformer不仅能够明显地提升翻译质量，还为许多NLP任务提供了新的结构。虽然原文写得很清楚，但实际上大家普遍反映很难正确地实现。 所以我们为此文章写了篇注解文档，并给出了一行行实现的Transformer的代码。本文档删除了原...
全网最强ViT (Vision Transformer)原理及代码解析
fulva的专栏
10-30 2万+
今天我们来详细了解一下Vision Transformer。基于timm的代码。 1. Patch Embedding Transformer原本是用来做NLP的工作的，所以ViT的首要任务是将图转换成词的结构，这里采取的方法是如上图左下角所示，将图片分割成小块，每个小块就相当于句子里的一个词。这里把每个小块称作Patch，而Patch Embedding就是把每个Patch再经过一个全连接网络压缩成一定维度的向量。 这里是VisionTransformer源代码中关于Patch Em...
scrapy爬虫时iseq = ((self.normkey(k), self.normvalue(v)) for k, v in seq)
baidu_31067827的博客
10-16 349
iseq = ((self.normkey(k), self.normvalue(v)) for k, v in seq) 1、检查headers：须为dict，不能为list 参考回答：https://stackoverflow.com/questions/36273347/scraping-google-analytics-by-scrapy ...
评论
成就一亿技术人!
拼手气红包6.0元
还能输入1000个字符
 
红包 添加红包
表情包 插入表情
表情包 代码片
 条评论被折叠 查看
添加红包

请填写红包祝福语或标题

个

红包个数最小为10个

元

红包金额最低5元

当前余额3.43元 前往充值 >
需支付：10.00元
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付元
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明：

1.余额是钱包充值的虚拟货币，按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载，可以购买VIP、付费专栏及课程。

余额充值