-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathatom.xml
348 lines (330 loc) · 15.3 KB
/
atom.xml
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
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
<?xml version="1.0" encoding="utf-8"?>
<feed xmlns="http://www.w3.org/2005/Atom">
<id>https://stedylan.github.io</id>
<title>Gridea</title>
<updated>2021-05-25T08:26:20.195Z</updated>
<generator>https://github.com/jpmonette/feed</generator>
<link rel="alternate" href="https://stedylan.github.io"/>
<link rel="self" href="https://stedylan.github.io/atom.xml"/>
<subtitle>温故而知新</subtitle>
<logo>https://stedylan.github.io/images/avatar.png</logo>
<icon>https://stedylan.github.io/favicon.ico</icon>
<rights>All rights reserved 2021, Gridea</rights>
<entry>
<title type="html"><![CDATA[pytorch]]></title>
<id>https://stedylan.github.io/post/pytorch初学图像分类器/</id>
<link href="https://stedylan.github.io/post/pytorch初学图像分类器/">
</link>
<updated>2021-12-11T16:00:00.000Z</updated>
<content type="html"><![CDATA[<h1 id="训练一个图像分类器">训练一个图像分类器</h1>
<p>1.使用torchvision加载和正则化CIFAR10训练集和测试集</p>
<p>2.构造一个CNN</p>
<p>3.构造损失函数</p>
<p>4.训练网络</p>
<p>5.测试网络</p>
<h2 id="1-加载和正规化cifar10">1. 加载和正规化CIFAR10</h2>
<pre><code class="language-python">import torch
import torchvision
import torchvision.transforms as transforms
</code></pre>
<pre><code class="language-python">transform = transforms.Compose(
[transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])
trainset = torchvision.datasets.CIFAR10(root='./data', train=True, download=True, transform=transform)
trainloader = torch.utils.data.DataLoader(trainset, batch_size=4, shuffle=True, num_workers=2)
testset = torchvision.datasets.CIFAR10(root='./data', train=False,
download=True, transform=transform)
testloader = torch.utils.data.DataLoader(testset, batch_size=4,
shuffle=False, num_workers=2)
classes = ('plane', 'car', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship', 'truck')
</code></pre>
<pre><code>Files already downloaded and verified
Files already downloaded and verified
</code></pre>
<pre><code class="language-python">import matplotlib.pyplot as plt
import numpy as np
def imshow(img):
img = img / 2 + 0.5
npimg = img.numpy()
plt.imshow(np.transpose(npimg, (1, 2, 0)))
plt.show()
dataiter = iter(trainloader)
images, labels = dataiter.next()
imshow(torchvision.utils.make_grid(images))
print(' '.join('%5s' % classes[labels[j]] for j in range(4)))
</code></pre>
<figure data-type="image" tabindex="1"><img src="https://img-blog.csdnimg.cn/20201222101039601.png#pic_center" alt="在这里插入图片描述" loading="lazy"></figure>
<pre><code>truck car bird ship
</code></pre>
<h2 id="2-定义一个卷积神经网络">2. 定义一个卷积神经网络</h2>
<pre><code class="language-python">import torch.nn as nn
import torch.nn.functional as F
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
class Net(nn.Module):
def __init__(self):
super(Net, self).__init__()
self.conv1 = nn.Conv2d(3, 6, 5)
self.pool = nn.MaxPool2d(2, 2)
self.conv2 = nn.Conv2d(6, 16, 5)
self.fc1 = nn.Linear(16 * 5 * 5, 120)
self.fc2 = nn.Linear(120, 84)
self.fc3 = nn.Linear(84, 10)
def forward(self, x):
x = self.pool(F.relu(self.conv1(x)))
x = self.pool(F.relu(self.conv2(x)))
x = x.view(-1, 16 * 5 * 5)
x = F.relu(self.fc1(x))
x = F.relu(self.fc2(x))
x = self.fc3(x)
return x
net = Net()
net.to(device)
</code></pre>
<pre><code>Net(
(conv1): Conv2d(3, 6, kernel_size=(5, 5), stride=(1, 1))
(pool): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(conv2): Conv2d(6, 16, kernel_size=(5, 5), stride=(1, 1))
(fc1): Linear(in_features=400, out_features=120, bias=True)
(fc2): Linear(in_features=120, out_features=84, bias=True)
(fc3): Linear(in_features=84, out_features=10, bias=True)
)
</code></pre>
<h2 id="3-定义损失函数和优化器">3. 定义损失函数和优化器</h2>
<pre><code class="language-python">import torch.optim as optim
criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(net.parameters(), lr=0.001, momentum=0.9) # momentum为动量，解决局部最优解问题
</code></pre>
<h2 id="4-训练cnn">4. 训练CNN</h2>
<pre><code class="language-python">
for epoch in range(2):
running_loss = 0.0
for i, data in enumerate(trainloader, 0):
# inputs, labels = data # 使用cpu训练
inputs, labels = data[0].to(device), data[1].to(device) # 使用gpu训练
optimizer.zero_grad()
outputs = net(inputs)
loss = criterion(outputs, labels)
loss.backward()
optimizer.step()
running_loss += loss.item()
if i % 2000 == 1999: # 每两千次，计算一下loss的总和，看总体的降低
print('[%d, %5d] loss: %.3f' % (epoch + 1, i + 1, running_loss / 2000))
running_loss = 0.0
print('Finished Training')
</code></pre>
<pre><code>[1, 2000] loss: 2.139
[1, 4000] loss: 1.818
[1, 6000] loss: 1.678
[1, 8000] loss: 1.575
[1, 10000] loss: 1.523
[1, 12000] loss: 1.490
[2, 2000] loss: 1.417
[2, 4000] loss: 1.345
[2, 6000] loss: 1.356
[2, 8000] loss: 1.314
[2, 10000] loss: 1.291
[2, 12000] loss: 1.279
Finished Training
</code></pre>
<pre><code class="language-python">PATH = './cifar_net.pth'
torch.save(net.state_dict(), PATH)
</code></pre>
<h2 id="5-测试网络">5. 测试网络</h2>
<pre><code class="language-python">dataiter = iter(testloader) # dataload迭代器
images, labels = dataiter.next()
# print images
imshow(torchvision.utils.make_grid(images))
print('GroundTruth: ', ' '.join('%5s' % classes[labels[j]] for j in range(4)))
</code></pre>
<figure data-type="image" tabindex="2"><img src="https://img-blog.csdnimg.cn/20201222101054704.png#pic_center" alt="在这里插入图片描述" loading="lazy"></figure>
<pre><code>GroundTruth: cat ship ship plane
</code></pre>
<pre><code class="language-python">net = Net()
net.load_state_dict(torch.load(PATH))
outputs = net(images)
</code></pre>
<pre><code class="language-python">_, predicted = torch.max(outputs, 1) # dim = 1, 将维度压缩为1
print(outputs) # 每一行代表一个图片的输出值
print(predicted) # 每张图片最大输出值的下边
print('Predicted: ', ' '.join('%5s' % classes[predicted[j]]
for j in range(4)))
</code></pre>
<pre><code>tensor([[-2.5090, -3.2349, 1.7301, 2.9694, 3.0115, 2.8149, 3.4539, 0.7578,
-3.6328, -3.4352],
[-0.8140, -4.5701, 2.3966, 2.8336, -0.1575, 5.3245, -0.4004, 1.8420,
-3.9213, -2.3566],
[ 1.6518, 0.1281, 2.1819, -0.2296, 4.1228, -0.2814, 1.4106, -0.8388,
-2.4360, -3.0393],
[-1.2521, -3.1468, 0.3812, 0.9961, 4.1565, 1.7851, 0.2695, 6.6411,
-5.3557, -1.3407]])
tensor([6, 5, 4, 7])
Predicted: frog dog deer horse
</code></pre>
<pre><code class="language-python">correct = 0
total = 0
with torch.no_grad(): # 预测的话不需要计算gradient
for data in testloader:
images, labels = data
outputs = net(images)
_, predicted = torch.max(outputs.data, 1)
total += labels.size(0)
correct += (predicted == labels).sum().item()
print('Accuracy of the network on the 10000 test images: %d %%' % (
100 * correct / total))
</code></pre>
<pre><code>Accuracy of the network on the 10000 test images: 56 %
</code></pre>
]]></content>
</entry>
<entry>
<title type="html"><![CDATA[DewarpNet]]></title>
<id>https://stedylan.github.io/post/README/</id>
<link href="https://stedylan.github.io/post/README/">
</link>
<updated>2020-12-11T16:00:00.000Z</updated>
<content type="html"><![CDATA[<h1 id="dewarpnet">DewarpNet</h1>
<p>This repository contains the codes for <a href="https://www3.cs.stonybrook.edu/~cvl/projects/dewarpnet/storage/paper.pdf"><strong>DewarpNet</strong></a> training.</p>
<h3 id="recent-updates">Recent Updates</h3>
<ul>
<li><strong>[May, 2020]</strong> Added evaluation images and an important note about Matlab SSIM.</li>
<li><strong>[Dec, 2020]</strong> Added OCR evaluation details.</li>
</ul>
<h3 id="training">Training</h3>
<ul>
<li>Prepare Data: <code>train.txt</code> & <code>val.txt</code>. Contents should be like:</li>
</ul>
<pre><code>1/824_8-cp_Page_0503-7Ns0001
1/824_1-cp_Page_0504-2Cw0001
</code></pre>
<ul>
<li>Train Shape Network:<br>
<code>python trainwc.py --arch unetnc --data_path ./data/DewarpNet/doc3d/ --batch_size 50 --tboard</code></li>
<li>Train Texture Mapping Network:<br>
<code>python trainbm.py --arch dnetccnl --img_rows 128 --img_cols 128 --img_norm --n_epoch 250 --batch_size 50 --l_rate 0.0001 --tboard --data_path ./DewarpNet/doc3d</code></li>
</ul>
<h3 id="inference">Inference:</h3>
<ul>
<li>Run:<br>
<code>python infer.py --wc_model_path ./eval/models/unetnc_doc3d.pkl --bm_model_path ./eval/models/dnetccnl_doc3d.pkl --show</code></li>
</ul>
<h3 id="evaluation-image-metrics">Evaluation (Image Metrics):</h3>
<ul>
<li>
<p>We use the same evaluation code as <a href="https://www3.cs.stonybrook.edu/~cvl/docunet.html">DocUNet</a>.<br>
To reproduce the quantitative results reported in the paper use the images available <a href="https://drive.google.com/drive/folders/1aPfQHGrGxpuIbYLONydbSkGNygRX2z2P?usp=sharing">here</a>.</p>
</li>
<li>
<p><strong>[Important note about Matlab version]</strong> We noticed that Matlab 2020a uses a different SSIM implementation which gives a better MS-SSIM score (0.5623). Whereas we have used Matlab 2018b. Please compare the scores according to your Matlab version.</p>
</li>
</ul>
<h3 id="evaluation-ocr-metrics">Evaluation (OCR Metrics):</h3>
<ul>
<li>The 25 images used for OCR evaluation is <code>/eval/ocr_eval/ocr_files.txt</code></li>
<li>The corresponding ground-truth text is given in <code>/eval/ocr_eval/tess_gt.json</code></li>
<li>For the OCR errors reported in the paper we had used cv2.blur as pre-processing which gives higher error in all the cases. For convenience, we provide the updated numbers (without using blur) in the following table:</li>
</ul>
<table>
<thead>
<tr>
<th style="text-align:center">Method</th>
<th style="text-align:center">ED</th>
<th style="text-align:center">CER</th>
<th style="text-align:center">ED (no blur)</th>
<th style="text-align:center">CER (no blur)</th>
</tr>
</thead>
<tbody>
<tr>
<td style="text-align:center">DocUNet</td>
<td style="text-align:center">1975.86</td>
<td style="text-align:center">0.4656(0.263)</td>
<td style="text-align:center">1671.80</td>
<td style="text-align:center">0.403 (0.256)</td>
</tr>
<tr>
<td style="text-align:center">DocUNet on Doc3D</td>
<td style="text-align:center">1684.34</td>
<td style="text-align:center">0.3955 (0.272)</td>
<td style="text-align:center">1296.00</td>
<td style="text-align:center">0.294 (0.235)</td>
</tr>
<tr>
<td style="text-align:center">DewarpNet</td>
<td style="text-align:center">1288.60</td>
<td style="text-align:center">0.3136 (0.248)</td>
<td style="text-align:center">1007.28</td>
<td style="text-align:center">0.249 (0.236)</td>
</tr>
<tr>
<td style="text-align:center">DewarpNet (ref)</td>
<td style="text-align:center">1114.40</td>
<td style="text-align:center">0.2692 (0.234)</td>
<td style="text-align:center">812.48</td>
<td style="text-align:center">0.204 (0.228)</td>
</tr>
</tbody>
</table>
<ul>
<li>We had used the Tesseract (v4.1.0) default configuration for evaluation with PyTesseract (v0.2.6).</li>
</ul>
<h3 id="models">Models:</h3>
<ul>
<li>Pre-trained models are available <a href="https://drive.google.com/file/d/1hJKCb4eF1AJih_dhZOJSF5VR-ZtRNaap/view?usp=sharing">here</a>. These models are captured prior to end-to-end training, thus won't give you the end-to-end results reported in Table 2 of the paper. Use the images provided above to get the exact numbers as Table 2.</li>
</ul>
<h3 id="dataset">Dataset:</h3>
<ul>
<li>The <em>doc3D dataset</em> can be downloaded using the scripts <a href="https://github.com/cvlab-stonybrook/doc3D-dataset">here</a>.</li>
</ul>
<h3 id="more-stuff">More Stuff:</h3>
<ul>
<li><a href="https://sagniklp.github.io/dewarpnet-demo/">Demo</a></li>
<li><a href="https://www3.cs.stonybrook.edu/~cvl/projects/dewarpnet/">Project Page</a></li>
<li><a href="https://github.com/sagniklp/doc3D-renderer">Doc3D Rendering Codes</a></li>
</ul>
<h3 id="citation">Citation:</h3>
<p>If you use the dataset or this code, please consider citing our work-</p>
<pre><code>@inproceedings{SagnikKeICCV2019,
Author = {Sagnik Das*, Ke Ma*, Zhixin Shu, Dimitris Samaras, Roy Shilkrot},
Booktitle = {Proceedings of International Conference on Computer Vision},
Title = {DewarpNet: Single-Image Document Unwarping With Stacked 3D and 2D Regression Networks},
Year = {2019}}
</code></pre>
<h4 id="acknowledgements">Acknowledgements:</h4>
<ul>
<li>These codes are heavily structured on <a href="https://github.com/meetshah1995/pytorch-semseg">pytorch-semseg</a>.</li>
</ul>
]]></content>
</entry>
<entry>
<title type="html"><![CDATA[Hello Gridea]]></title>
<id>https://stedylan.github.io/post/hello-gridea/</id>
<link href="https://stedylan.github.io/post/hello-gridea/">
</link>
<updated>2018-12-11T16:00:00.000Z</updated>
<summary type="html"><![CDATA[<p>👏 欢迎使用 <strong>Gridea</strong> ！<br>
✍️ <strong>Gridea</strong> 一个静态博客写作客户端。你可以用它来记录你的生活、心情、知识、笔记、创意... ...</p>
]]></summary>
<content type="html"><![CDATA[<p>👏 欢迎使用 <strong>Gridea</strong> ！<br>
✍️ <strong>Gridea</strong> 一个静态博客写作客户端。你可以用它来记录你的生活、心情、知识、笔记、创意... ...</p>
<!-- more -->
<p><a href="https://github.com/getgridea/gridea">Github</a><br>
<a href="https://gridea.dev/">Gridea 主页</a><br>
<a href="http://fehey.com/">示例网站</a></p>
<h2 id="特性">特性👇</h2>
<p>📝 你可以使用最酷的 <strong>Markdown</strong> 语法，进行快速创作</p>
<p>🌉 你可以给文章配上精美的封面图和在文章任意位置插入图片</p>
<p>🏷️ 你可以对文章进行标签分组</p>
<p>📋 你可以自定义菜单，甚至可以创建外部链接菜单</p>
<p>💻 你可以在 <strong>Windows</strong>，<strong>MacOS</strong> 或 <strong>Linux</strong> 设备上使用此客户端</p>
<p>🌎 你可以使用 <strong>𝖦𝗂𝗍𝗁𝗎𝖻 𝖯𝖺𝗀𝖾𝗌</strong> 或 <strong>Coding Pages</strong> 向世界展示，未来将支持更多平台</p>
<p>💬 你可以进行简单的配置，接入 <a href="https://github.com/gitalk/gitalk">Gitalk</a> 或 <a href="https://github.com/SukkaW/DisqusJS">DisqusJS</a> 评论系统</p>
<p>🇬🇧 你可以使用<strong>中文简体</strong>或<strong>英语</strong></p>
<p>🌁 你可以任意使用应用内默认主题或任意第三方主题，强大的主题自定义能力</p>
<p>🖥 你可以自定义源文件夹，利用 OneDrive、百度网盘、iCloud、Dropbox 等进行多设备同步</p>
<p>🌱 当然 <strong>Gridea</strong> 还很年轻，有很多不足，但请相信，它会不停向前 🏃</p>
<p>未来，它一定会成为你离不开的伙伴</p>
<p>尽情发挥你的才华吧！</p>
<p>😘 Enjoy~</p>
]]></content>
</entry>
</feed>