深度學習與Pytorch入門實戰（七）Visdom可視化工具【數字識別實例】
https://www.cnblogs.com/douzujun/p/13324435.html
【深度學習-pytorch-番外篇】如何使用Tensorboard可視化Pytorch訓練結果
https://blog.csdn.net/Kefenggewu_/article/details/118292747?spm=1001.2101.3001.6661.1&utm_medium=distribute.pc_relevant_t0.none-task-blog-2%7Edefault%7ECTRLIST%7ERate-1.pc_relevant_aa&depth_1-utm_source=distribute.pc_relevant_t0.none-task-blog-2%7Edefault%7ECTRLIST%7ERate-1.pc_relevant_aa&utm_relevant_index=1
Pytorch訓練可視化(TensorboardX)
https://zhuanlan.zhihu.com/p/54947519
[官方總結] tensorboardX 使用教程
https://blog.csdn.net/qq_39575835/article/details/89160828
torch.nn.conv3d理解
https://blog.csdn.net/weixin_42769131/article/details/104826953
支持PyTorch的einops張量操作神器用法示例詳解
https://www.jb51.net/article/226979.htm
https://www.cnblogs.com/c-chenbin/p/15375637.html
詳解PyTorch中的ModuleList和Sequential
https://zhuanlan.zhihu.com/p/75206669
Pytorch tf.nn.functional.softmax(x,dim = -1)對參數dim的理解
https://blog.csdn.net/weixin_44317740/article/details/107373202
Numpy中transpose()函數的可視化理解
https://zhuanlan.zhihu.com/p/61203757
對pytorch中x = x.view(x.size(0), -1) 的理解說明
https://www.jb51.net/article/206748.htm
https://mapengsen.blog.csdn.net/article/details/115214078
Python3.x：python: extend (擴展) 與 append (追加) 的區別
https://www.cnblogs.com/lizm166/p/8232733.html
nn.linear()函數
https://blog.csdn.net/daydaydreamer/article/details/102638624
torch.nn.Embedding()函數解讀
https://blog.csdn.net/qq_40178291/article/details/100658867
pytorch LayerNorm參數的用法及計算過程
https://www.jb51.net/article/213383.htm
nn.LayerNorm的參數
https://blog.csdn.net/qq_45893319/article/details/122327931
Pytorch數據預處理：transforms的使用方法
https://zhuanlan.zhihu.com/p/130985895
transforms.Resize(256)是按照比例把圖像最小的一個邊長放縮到256，另一邊按照相同比例放縮。
transforms.RandomResizedCrop(224,scale=(0.5,1.0))是把圖像按照中心隨機切割成224正方形大小的圖片。
# transforms.Resize([h, w])
pytorch torch.manual_seed()用法
https://www.cnblogs.com/dychen/p/13920000.html
normal_kernel_cpu“ not implemented for ‘Long‘
https://blog.csdn.net/YUwoshijiantian/article/details/115219686
在0后面加個 .
pytorch讀取本地的mnist數據集【實測成功】
https://blog.csdn.net/weixin_41529093/article/details/111354381?spm=1001.2101.3001.6650.1&utm_medium=distribute.pc_relevant.none-task-blog-2%7Edefault%7ECTRLIST%7ERate-1.pc_relevant_default&depth_1-utm_source=distribute.pc_relevant.none-task-blog-2%7Edefault%7ECTRLIST%7ERate-1.pc_relevant_default&utm_relevant_index=2
需要導入：import gzip ，取消類名括號內的參數
Pytorch MNIST直接離線加載二進制文件到pytorch
https://blog.csdn.net/caomin1hao/article/details/108522205
pytorch-構建自己的dataset類
https://blog.csdn.net/l641208111/article/details/113838584
https://zhuanlan.zhihu.com/p/159484351
https://www.jianshu.com/p/4818a1a4b5bd
https://www.cnblogs.com/popodynasty/p/15170266.html
https://blog.csdn.net/zw__chen/article/details/82806900
https://blog.csdn.net/sinat_42239797/article/details/90641659?spm=1001.2101.3001.6650.1&utm_medium=distribute.pc_relevant.none-task-blog-2%7Edefault%7ECTRLIST%7ERate-1.pc_relevant_default&depth_1-utm_source=distribute.pc_relevant.none-task-blog-2%7Edefault%7ECTRLIST%7ERate-1.pc_relevant_default&utm_relevant_index=2
https://blog.csdn.net/qq_34107425/article/details/104097402?spm=1001.2101.3001.6650.1&utm_medium=distribute.pc_relevant.none-task-blog-2%7Edefault%7ECTRLIST%7ERate-1.pc_relevant_default&depth_1-utm_source=distribute.pc_relevant.none-task-blog-2%7Edefault%7ECTRLIST%7ERate-1.pc_relevant_default&utm_relevant_index=2
pytorch如何顯示數據圖像，以及標簽。TypeError: img should be PIL Image. Got ＜class ‘numpy.ndarray‘＞
所以我們要在轉換中先轉換為PIL格式。
transforms.ToPILImage()
https://blog.csdn.net/wacebb/article/details/108003306
https://blog.csdn.net/qq_40178291/article/details/101108327
PIL.JpegImagePlugin.JpegImageFile與numpy.ndarray的相互轉換
https://blog.csdn.net/hua_you_qiang/article/details/118683134?utm_medium=distribute.pc_aggpage_search_result.none-task-blog-2~aggregatepage~first_rank_ecpm_v1~rank_v31_ecpm-1-118683134.pc_agg_new_rank&utm_term=ndarray%E8%BD%ACjpegimagefile&spm=1000.2123.3001.4430
np.ndarray與torch.Tensor之間的轉化 （圖像的區別）
https://blog.csdn.net/weixin_45508265/article/details/119040774?utm_medium=distribute.pc_relevant.none-task-blog-2~default~baidujs_title~default-0.pc_relevant_default&spm=1001.2101.3001.4242.1&utm_relevant_index=3
【學習筆記】pytorch中squeeze()和unsqueeze()函數介紹
https://blog.csdn.net/flysky_jay/article/details/81607289
高版本pytorch出現IndexError: invalid index of a 0-dim tensor.問題解決辦法
https://blog.csdn.net/qq_31511955/article/details/111829976?spm=1001.2101.3001.6661.1&utm_medium=distribute.pc_relevant_t0.none-task-blog-2%7Edefault%7ECTRLIST%7ERate-1.pc_relevant_default&depth_1-utm_source=distribute.pc_relevant_t0.none-task-blog-2%7Edefault%7ECTRLIST%7ERate-1.pc_relevant_default&utm_relevant_index=1
NameError:name ‘xrange’ is not defined
https://www.cnblogs.com/hdk1993/p/8893991.html
PyTorch:expected scalar type Float but found Double
https://chenlinwei.blog.csdn.net/article/details/109725458?spm=1001.2101.3001.6661.1&utm_medium=distribute.pc_relevant_t0.none-task-blog-2%7Edefault%7ECTRLIST%7ERate-1.pc_relevant_paycolumn_v3&depth_1-utm_source=distribute.pc_relevant_t0.none-task-blog-2%7Edefault%7ECTRLIST%7ERate-1.pc_relevant_paycolumn_v3&utm_relevant_index=1
這個問題很明顯就是網絡內的參數類型不同；
修改：
在前面添加：
torch.set_default_tensor_type(torch.DoubleTensor)
或者，在運行網絡前添加：
net = net.double()
https://blog.csdn.net/weixin_41514525/article/details/110129655
多GPU訓練相關問題
pytorch錯誤：Missing key(s) in state_dict、Unexpected key(s) in state_dict解決
https://www.cnblogs.com/zhengbiqing/p/10434704.html
Missing key(s) in state_dict: “module.features.0.0.weight
https://blog.csdn.net/qq_35037684/article/details/115109416
[python][pytorch]多GPU下的模型保存與加載
https://www.cnblogs.com/wildkid1024/p/13025352.html
https://github.com/bearpaw/pytorch-classification/issues/27
【Pytorch多GPU訓練錯誤】AttributeError: ‘DataParallel’ object has no attribute ‘xxxx
https://blog.csdn.net/weixin_41990278/article/details/105127101
【PyTorch問題】can‘t convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy...略
https://blog.csdn.net/xiaoxiao_ziteng/article/details/115432973?spm=1001.2101.3001.6650.1&utm_medium=distribute.pc_relevant.none-task-blog-2%7Edefault%7ECTRLIST%7ERate-1.topblog&depth_1-utm_source=distribute.pc_relevant.none-task-blog-2%7Edefault%7ECTRLIST%7ERate-1.topblog&utm_relevant_index=2
https://blog.csdn.net/wacebb/article/details/114652811
在我們想把 GPU tensor 轉換成 Numpy 變量的時候，需要先將 tensor 轉換到 CPU 去，因為 Numpy 是 CPU-only 的。
訓練模型的時候， Warning: NaN or Inf found in input tensor 解決辦法
https://blog.csdn.net/qq_38284961/article/details/102935800
https://blog.csdn.net/JSLS_Hf/article/details/81743045?spm=1001.2101.3001.6650.1&utm_medium=distribute.pc_relevant.none-task-blog-2%7Edefault%7ECTRLIST%7ERate-1.pc_relevant_paycolumn_v3&depth_1-utm_source=distribute.pc_relevant.none-task-blog-2%7Edefault%7ECTRLIST%7ERate-1.pc_relevant_paycolumn_v3&utm_relevant_index=2
https://blog.csdn.net/weixin_41278720/article/details/80778640?spm=1001.2101.3001.6650.1&utm_medium=distribute.pc_relevant.none-task-blog-2%7Edefault%7ECTRLIST%7ERate-1.pc_relevant_paycolumn_v3&depth_1-utm_source=distribute.pc_relevant.none-task-blog-2%7Edefault%7ECTRLIST%7ERate-1.pc_relevant_paycolumn_v3&utm_relevant_index=2
Detected call of `lr_scheduler.step()` before `optimizer.step()`.
https://blog.csdn.net/weixin_38314865/article/details/103937717
https://blog.csdn.net/qq_41166909/article/details/122531321?spm=1001.2101.3001.6650.1&utm_medium=distribute.pc_relevant.none-task-blog-2%7Edefault%7EBlogCommendFromBaidu%7EHighlightScore-1.queryctrv2&depth_1-utm_source=distribute.pc_relevant.none-task-blog-2%7Edefault%7EBlogCommendFromBaidu%7EHighlightScore-1.queryctrv2&utm_relevant_index=2
應該把lr_scheduler.step()放在每次epoch訓練完成之后
https://zhuanlan.zhihu.com/p/136902153
model.train()與model.eval()的用法
https://www.cnblogs.com/elitphil/p/15532447.html
https://blog.csdn.net/kking_edc/article/details/104663305
model.train():
在使用pytorch構建神經網絡的時候，訓練過程中會在程序上方添加一句model.train()，作用是啟用batch normalization和drop out。
model.eval():
測試過程中會使用model.eval()，這時神經網絡會沿用batch normalization的值，並不使用drop out。
pycharm代碼塊左移、右移
在使用pycharm編寫代碼時，會遇到代碼塊左移右移的操作
1.代碼快右移：選中多行代碼，tab鍵縮進，一次縮進4個字符
2.代碼塊左移：選中多行代碼，shift+tab，一次左移4個字符
Pytorch-對於pytorch下載CIFAR10數據集很慢或卡住的解決方法
https://blog.csdn.net/qq_28790663/article/details/115032503
Ubuntu16.04下Python程序出現錯誤qt.qpa.plugin: Could not load the Qt platform plugin xcb解決方法
https://blog.csdn.net/zhanghm1995/article/details/106474505
https://www.jb51.net/article/193024.htm
https://blog.csdn.net/agonysome/article/details/108985079
QObject::moveToThread: Current thread(...) is not the object`s thread. Cannot move to target thread(
https://chowdera.com/2022/02/202202100555216474.html
https://blog.csdn.net/qq_29750461/article/details/109720034?spm=1001.2101.3001.6650.9&utm_medium=distribute.pc_relevant.none-task-blog-2%7Edefault%7ECTRLIST%7ERate-9.pc_relevant_paycolumn_v3&depth_1-utm_source=distribute.pc_relevant.none-task-blog-2%7Edefault%7ECTRLIST%7ERate-9.pc_relevant_paycolumn_v3&utm_relevant_index=12
https://blog.csdn.net/weixin_42326323/article/details/99309231?spm=1001.2101.3001.6650.3&utm_medium=distribute.pc_relevant.none-task-blog-2%7Edefault%7EBlogCommendFromBaidu%7EHighlightScore-3.queryctrv2&depth_1-utm_source=distribute.pc_relevant.none-task-blog-2%7Edefault%7EBlogCommendFromBaidu%7EHighlightScore-3.queryctrv2&utm_relevant_index=6
https://blog.csdn.net/weixin_38082364/article/details/89234765
Pytorch-18種經典的損失函數
https://blog.csdn.net/weixin_43687366/article/details/107927693
https://www.jianshu.com/p/4812f381a24a
https://blog.csdn.net/tototuzuoquan/article/details/113777788
python批量將視頻轉化為圖片
https://www.cnblogs.com/stupidwf/p/13338218.html
https://www.pythonheidong.com/blog/article/1163380/3df3c70fa0d484c62dd4/
CV2逐步學習-1.imread()詳解+cvtColor()顏色空間轉換
https://blog.csdn.net/sunjintaoxxx/article/details/121262553
img=cv2.imread('aima.jpg')
img_rgb=cv2.cvtColor(img,cv2.COLOR_BGR2RGB)
Softmax函數和Sigmoid函數的區別與聯系
https://zhuanlan.zhihu.com/p/356976844
EMD(earth mover's distances)距離
https://zhuanlan.zhihu.com/p/145739750
【pytorch系列】 with torch.no_grad():用法詳解
https://blog.csdn.net/sazass/article/details/116668755
pandas to_csv() 索引列表第一行是0的問題
https://blog.csdn.net/Ly_Word/article/details/116152644
pandas中Dataframe選取指定行和列或刪除含有指定數值的行或者列
https://blog.csdn.net/wf592523813/article/details/96278289
數據框DataFrame和列表List相互轉換
https://www.cnblogs.com/xiaodangdang/p/12098137.html
json轉化為dataframe 和dataframe轉化為json
https://blog.csdn.net/sslfk/article/details/122824057?spm=1001.2101.3001.6650.1&utm_medium=distribute.pc_relevant.none-task-blog-2%7Edefault%7ECTRLIST%7Edefault-1.pc_relevant_default&depth_1-utm_source=distribute.pc_relevant.none-task-blog-2%7Edefault%7ECTRLIST%7Edefault-1.pc_relevant_default&utm_relevant_index=2
Python3 * 和 ** 運算符
https://blog.csdn.net/yilovexing/article/details/80577510
python中 // 和 / 和 %
https://blog.csdn.net/qq_29566629/article/details/95374971?spm=1001.2101.3001.6661.1&utm_medium=distribute.pc_relevant_t0.none-task-blog-2%7Edefault%7ECTRLIST%7ERate-1.pc_relevant_aa&depth_1-utm_source=distribute.pc_relevant_t0.none-task-blog-2%7Edefault%7ECTRLIST%7ERate-1.pc_relevant_aa&utm_relevant_index=1
“/”，這是傳統的除法，3/2=1.5
“//”，在python中，這個叫“地板除”，3//2=1
“%”，這個是取模操作，也就是區余數，4%2=0，5%2=1
序列解包
https://blog.csdn.net/yilovexing/article/details/80576788
python內置函數：enumerate用法總結
https://blog.csdn.net/IAMoldpan/article/details/78487809
enumerate(iterable, start=0)
第一個參數為可迭代的數據，比如python中的list。第二個參數為該函數打印標號的初始值，默認從0開始打印，該函數返回一個enumerate類型的數據。
*args 和 **kwargs 主要用於函數定義。
https://blog.csdn.net/yilovexing/article/details/80577510
*args 與 **kwargs 的區別，兩者都是 python 中的可變參數：
*args 表示任何多個無名參數，它本質是一個 tuple
**kwargs 表示關鍵字參數，它本質上是一個 dict
如果同時使用 *args 和 **kwargs 時，必須 *args 參數列要在 **kwargs 之前。
python -- 定義函數 def 后面的 -＞，：表示的含義
https://blog.csdn.net/qq_40913465/article/details/108407867
設置python搜索路徑的幾種方法
https://blog.csdn.net/phy12321/article/details/104137387?spm=1001.2101.3001.6650.1&utm_medium=distribute.pc_relevant.none-task-blog-2%7Edefault%7ECTRLIST%7ERate-1.pc_relevant_default&depth_1-utm_source=distribute.pc_relevant.none-task-blog-2%7Edefault%7ECTRLIST%7ERate-1.pc_relevant_default&utm_relevant_index=2
python臨時添加當前工作路徑
export PYTHONPATH=$PYTHONPATH:./
這樣找模塊就方便多了
python之parser.add_argument()用法——命令行選項、參數和子命令解析器
https://blog.csdn.net/qq_34243930/article/details/106517985
【Python】python中argparse.add_argument中的action=‘store_true‘使用總結
https://blog.csdn.net/AugustMe/article/details/109593110
AttributeError: module 'cv2.cv2' has no attribute 'DualTVL1OpticalFlow_create'
https://www.codeleading.com/article/21272688801/
opencv4.0以下版本API接口“cv2.DualTVL1OpticalFlow_create()”，opencv4.0以上版本API接口使用“cv2.optflow.DualTVL1OpticalFlow_create()
pip install opencv_python==4.1.2.30
pip install opencv-contrib-python==4.1.2.30
AttributeError: module 'cv2.optflow' has no attribute 'DISOpticalFlow_create'
https://answers.opencv.org/question/212935/disoptical-flow-in-41/
# 3.4.4.19 #inst = cv2.optflow.createOptFlow_DIS(cv2.optflow.DISOPTICAL_FLOW_PRESET_MEDIUM) # 4.1 inst = cv2.DISOpticalFlow_create(cv2.DISOPTICAL_FLOW_PRESET_MEDIUM)
AttributeError: module 'cv2.optflow' has no attribute 'DISOpticalFlow_create'
Legacy autograd function with non-static forward method is deprecated.
https://zhuanlan.zhihu.com/p/355875710
記錄自己調試SSD過程(第一次寫，如有不對，請各位指正)
https://www.icode9.com/content-4-1021875.html
https://www.pythonheidong.com/blog/article/1412072/eb89ccea2f26626bda76/
https://blog.csdn.net/jacke121/article/details/116549423
ValueError: threshold must be numeric and non-NAN, try sys.maxsize for untruncated representation
https://blog.csdn.net/weixin_45752264/article/details/123406374
ModuleNotFoundError：No module named ‘pycocotools‘的解決辦法匯總
https://blog.csdn.net/qq_42672745/article/details/116358111
pip install pycocotools-windows
Python數據預處理之數據規范化
https://blog.csdn.net/weixin_46599926/article/details/123888199
error: ‘constexpr’ needed for in-class initialization of static data member ‘const double sba::Con2d
https://blog.csdn.net/wphkadn/article/details/88174109
const static 和 static const一樣，都不能在類內直接初始化非整形常量，可以修飾int，bool，char，但不能修飾其他類型（如double，float）
在c++11中，可以使用 constexpr static 或者 static constexpr 來修飾 非整形靜態成員常量。
也就是說，上面這一行出錯的代碼要改成
constexpr static double qScale = 1.0;
//或者 static constexpr double qScale = 1.0;