1.参考aaaaabin博主:https://blog.csdn.net/weixin_44490080/article/details/104283455
参考学习:https://blog.csdn.net/weixin_44490080/article/details/103925188
2.aaaaabin博主使用2018版本的分类例程,小改形成了这个demo,在原来基础上,我增加了:保存test结果的代码;文章中黄色字体为封装函数,代码段是封装函数的内容,大家把代码段内容封装好,就可以正常使用了。
3.遇到问题:图片直接从网络下载,保存为.png后缀,读取时,读取不了。解决:2345看图王,批量转换图片为png格式,猜测保存图片是我自己改的名字,可能本身并不是png格式,导致的错误。
4.完整代码百度云盘:
链接:https://pan.baidu.com/s/1_7y7tniFdWPyZHNnuE173A
提取码:25ri
复制这段内容后打开百度网盘手机App,操作更方便哦
read_dl_classifier ('pretrained_dl_classifier_compact.hdl', DLClassifierHandle) //读取网络模型
get_dl_classifier_param (DLClassifierHandle, 'image_width', DlImageWidth) //读取网络需要的图片大小
get_dl_classifier_param (DLClassifierHandle, 'image_height', DlImageHeight)
get_dl_classifier_param (DLClassifierHandle, 'image_num_channels', DlNumChannels) //读取网络需要的图片通道数
get_dl_classifier_param (DLClassifierHandle, 'image_range_min', DlRangeMin) //读取网络需要的图片灰度值范围
get_dl_classifier_param (DLClassifierHandle, 'image_range_max', DlRangeMax)
* 训练样本图就要做一个预处理,把每个样本图才处理成224*224分辨率、3通道、-127到128亮度级图像。
dev_close_window ()
WindowWidth:=800
WindowHeight:=600
dev_open_window (0, 0, WindowWidth, WindowHeight, 'black', WindowHandle)
*数据预处理
RawDataFolder := 'ImageData/'+['OK','NG']
read_dl_classifier_data_set (RawDataFolder, 'last_folder', RawImageFiles, Labels, LabelIndices, Classes)
PreprocessedFolder := 'preprocessedFolder'
OverwritePreprocessingFolder := true
*RemovePreprocessingAfterExample := true
file_exists (PreprocessedFolder, FileExists)
if (not FileExists or OverwritePreprocessingFolder)
if (FileExists)
remove_dir_recursively (PreprocessedFolder)
endif
make_dir (PreprocessedFolder)
for I := 0 to |Classes| - 1 by 1
make_dir (PreprocessedFolder + '/' + Classes[I])
endfor
parse_filename (RawImageFiles, BaseNames, Extensions, Directories)
ObjectFilesOut := PreprocessedFolder + '/' + Labels + '/' + BaseNames + '.hobj'
check_output_file_names_for_duplicates (RawImageFiles, ObjectFilesOut)
CheckOutputFiles := uniq(sort(ObjectFilesOut))
if (|CheckOutputFiles| != |RawImageFiles|)
SortedImageFiles := sort(ObjectFilesOut)
for I := 0 to |SortedImageFiles| - 1 by 1
if (SortedImageFiles[I] != CheckOutputFiles[I])
throw ('Error some file(s) have the same output filenames: ' + SortedImageFiles[I])
endif
endfor
endif
return ()
for I := 0 to |RawImageFiles| - 1 by 1
read_image (Image, RawImageFiles[I]) //读取样本文件
zoom_image_size (Image, Image, DlImageWidth, DlImageHeight, 'constant') //将图片缩放到网络model需求的大小
convert_image_type (Image, Image, 'real') //将图像的灰度缩放成网络model需求范围
RescaleRange:=(DlRangeMax - DlRangeMin)/255.0
scale_image (Image, Image, RescaleRange, DlRangeMin)
count_obj (Image, Number) //合成三通道图片
for Index := 1 to Number by 1
select_obj (Image, ObjectSelected, Index)
count_channels (ObjectSelected, Channel)
if (Channel != DlNumChannels) //如果图片不是三通道图,就需要将图像合成三通道图
compose3(ObjectSelected, ObjectSelected, ObjectSelected, ThreeChannelImage)
replace_obj (Image, ThreeChannelImage, Image, 1) //替换图元数组
endif
endfor
* 将预处理后的图像写入hobj文件
write_object (Image, ObjectFilesOut[I])
endfor
dev_clear_window ()
dev_disp_text ('图片预处理阶段完成!', 'window', 'top', 'left', 'black', [], [])
endif
read_dl_classifier_data_set (PreprocessedFolder, 'last_folder', ImageFiles, Labels, LabelsIndices, Classes)
*训练占比
TrainingPercent := 70
*验证占比
ValidationPercent := 15
*将数据分割成三大块,分别为:训练集(TrainingImages, TrainingLabels)、验证集(ValidationImages, ValidationLabels)、测试集(TestImages, TestLabels)
split_dl_classifier_data_set (ImageFiles, Labels, TrainingPercent, ValidationPercent, TrainingImages, TrainingLabels, ValidationImages, ValidationLabels, TestImages, TestLabels)
* This procedure divides the data set (images and ground truth labels)
* into three disjoint subsets: training, validation, and test.
* The number of images and labels in each subset is defined
* by the given percentages TrainingPercent and ValidationPercent.
* Each subset contains randomly distributed data,
* whereby the original ratio of class sizes is kept.
*
* Check the input parameters.
if (|ImageFiles| != |GroundTruthLabels|)
throw ('Please provide a label for every image file.')
endif
if (TrainingPercent < 0)
throw ('TrainingPercent must not be smaller than zero.')
endif
if (ValidationPercent < 0)
throw ('ValidationPercent must not be smaller than zero.')
endif
if (|ImageFiles| < 1)
throw ('ImageFiles must not be empty.')
endif
if (TrainingPercent + ValidationPercent > 100)
throw ('The sum of TrainingPercent and ValidationPercent must not be greater than 100.')
endif
*
* Set classes and data ratios.
TrainingRatio := TrainingPercent * 0.01
ValidationRatio := ValidationPercent * 0.01
*
* Prepare output tuples.
TrainingImages := []
TrainingLabels := []
ValidationImages := []
ValidationLabels := []
TestImages := []
TestLabels := []
*
* Loop through all unique classes and add data
* according to the specified percentages.
UniqueClasses := uniq(sort(GroundTruthLabels))
for ClassIndex := 0 to |UniqueClasses| - 1 by 1
* Select all images and ground truth labels with the class.
Class := UniqueClasses[ClassIndex]
ClassIndices := find(GroundTruthLabels,Class)
ImageFilesClass := ImageFiles[ClassIndices]
LabelsClass := gen_tuple_const(|ImageFilesClass|,Class)
* Shuffle the images in this class.
tuple_shuffle (ImageFilesClass, ImageFilesClass)
* Determine the boundaries of the respective selection.
IndexTrainingEnd := int(floor(|ImageFilesClass| * TrainingRatio)) - 1
IndexValidationEnd := int(floor(|ImageFilesClass| * (ValidationRatio + TrainingRatio))) - 1
* Add the respective images and labels.
TrainingImages := [TrainingImages,ImageFilesClass[0:IndexTrainingEnd]]
TrainingLabels := [TrainingLabels,LabelsClass[0:IndexTrainingEnd]]
ValidationImages := [ValidationImages,ImageFilesClass[IndexTrainingEnd + 1:IndexValidationEnd]]
ValidationLabels := [ValidationLabels,LabelsClass[IndexTrainingEnd + 1:IndexValidationEnd]]
TestImages := [TestImages,ImageFilesClass[IndexValidationEnd + 1:|ImageFilesClass| - 1]]
TestLabels := [TestLabels,LabelsClass[IndexValidationEnd + 1:|ImageFilesClass| - 1]]
endfor
*
* Shuffle the output.
tuple_shuffle ([0:|TrainingImages| - 1], TrainingSequence)
TrainingImages := TrainingImages[TrainingSequence]
TrainingLabels := TrainingLabels[TrainingSequence]
tuple_shuffle ([0:|ValidationImages| - 1], ValidationSequence)
ValidationImages := ValidationImages[ValidationSequence]
ValidationLabels := ValidationLabels[ValidationSequence]
tuple_shuffle ([0:|TestImages| - 1], TestSequence)
TestImages := TestImages[TestSequence]
TestLabels := TestLabels[TestSequence]
return ()
stop ()
**设置超参数**
*设置类别超参数
set_dl_classifier_param (DLClassifierHandle, 'classes', Classes)
BatchSize := 5
set_dl_classifier_param (DLClassifierHandle, 'batch_size', BatchSize)
try
*初始化网络模型
set_dl_classifier_param (DLClassifierHandle, 'runtime_init', 'immediately')
catch (Exception)
dev_disp_error_text (Exception)
ErrorAndAdviceText := 'An error occurred during runtime initialization.'
ErrorAndAdviceText := [ErrorAndAdviceText,'']
ErrorAndAdviceText := [ErrorAndAdviceText,'Error ' + Exception[0] + ': \'' + Exception[2] + '\'']
if (Exception[0] == 4104)
* In case of out of device memory we can give advice.
ErrorAndAdviceText := [ErrorAndAdviceText,'']
ErrorAndAdviceText := [ErrorAndAdviceText,'Install a GPU with more RAM or reduce the batch size.']
ErrorAndAdviceText := [ErrorAndAdviceText,'']
ErrorAndAdviceText := [ErrorAndAdviceText,'Note that changing the batch size will have an influence on the results.']
endif
dev_clear_window ()
* Display text with line breaks after 60 characters.
dev_disp_text (regexp_replace(ErrorAndAdviceText + ' ',['(.{0,60})\\s','replace_all'],'$1\n'), 'window', 'center', 'left', 'red', [], [])
return ()
stop ()
endtry
*学习率
InitialLearningRate := 0.001
set_dl_classifier_param (DLClassifierHandle, 'learning_rate', InitialLearningRate)
*学习率变化参数
LearningRateStepEveryNthEpoch := 50
LearningRateStepRatio := 0.1
*迭代次数
NumEpochs := 100
**训练分类器**
dev_clear_window ()
* 每次迭代的间隔,它会体现在学习过程中的图标上的‘x’轴
PlotIterationInterval := 1
*将训练好的网络模型序列化
FileName := 'classifier_minist.hdl'
*训练模型
train_fruit_classifier (DLClassifierHandle, FileName, NumEpochs, TrainingImages, TrainingLabels, ValidationImages, ValidationLabels, LearningRateStepEveryNthEpoch, LearningRateStepRatio, PlotIterationInterval, WindowHandle)
* For the plot during training,
* we need to concatenate some intermediate results.
TrainingErrors := []
ValidationErrors := []
LearningRates := []
Epochs := []
LossByIteration := []
get_dl_classifier_param (DLClassifierHandle, 'batch_size', BatchSize)
MinValidationError := 1
*
* Create a tuple that includes all the iterations
* where the plot should be computed (including the last ieration).
NumBatchesInEpoch := int(floor(|TrainingImages| / real(BatchSize)))
NumTotalIterations := (NumBatchesInEpoch * NumEpochs) - 1
PlottedIterations := round([NumBatchesInEpoch * [0:PlotEveryNthEpoch:NumEpochs - 1],NumTotalIterations])
*
* TrainSequence is used for easier indexing of the training data.
tuple_gen_sequence (0, |TrainingImages| - 1, 1, TrainSequence)
*
* Select a subset of the training data set
* in order to obtain a fast approximation
* of the training error during training (plotting).
SelectPercentageTrainingImages := 100
select_percentage_dl_classifier_data (TrainingImages, TrainingLabels, SelectPercentageTrainingImages, TrainingImagesSelected, TrainingLabelsSelected)
*
for Epoch := 0 to NumEpochs - 1 by 1
* In order to get randomness in each epoch,
* the training set is shuffled every epoch.
tuple_shuffle (TrainSequence, TrainSequence)
for Iteration := 0 to NumBatchesInEpoch - 1 by 1
* Select a batch from the training data set.
BatchStart := Iteration * BatchSize
BatchEnd := BatchStart + (BatchSize - 1)
BatchIndices := TrainSequence[BatchStart:BatchEnd]
BatchImageFiles := TrainingImages[BatchIndices]
BatchLabels := TrainingLabels[BatchIndices]
*
* Read the image of the current batch.
read_image (BatchImages, BatchImageFiles)
* Augment the images to get a better variety of training images.
GenParamName := 'mirror'
GenParamValue := 'rc'
augment_images (BatchImages, BatchImages, GenParamName, GenParamValue)
*
* Train the network with these images and ground truth labels.
train_dl_classifier_batch (BatchImages, DLClassifierHandle, BatchLabels, DLClassifierTrainResultHandle)
* You can access the current value of the loss function,
* which should decrease during the training.
get_dl_classifier_train_result (DLClassifierTrainResultHandle, 'loss', Loss)
* Store the loss in a tuple .
LossByIteration := [LossByIteration,Loss]
*
* In regular intervals, we want to evaluate
* how well our classifier performs.
CurrentIteration := int(Iteration + (NumBatchesInEpoch * Epoch))
if (sum(CurrentIteration [==] PlottedIterations))
* Plot the progress regularly.
* Evaluate the current classifier on the training and validation set.
apply_dl_classifier_batchwise (TrainingImagesSelected, DLClassifierHandle, TrainingDLClassifierResultIDs, TrainingPredictedLabels, TrainingConfidences)
apply_dl_classifier_batchwise (ValidationImages, DLClassifierHandle, ValidationDLClassifierResultIDs, ValidationPredictedLabels, ValidationConfidences)
* Evaluate the top-1 error on each dataset.
evaluate_dl_classifier (TrainingLabelsSelected, DLClassifierHandle, TrainingDLClassifierResultIDs, 'top1_error', 'global', TrainingTop1Error)
evaluate_dl_classifier (ValidationLabels, DLClassifierHandle, ValidationDLClassifierResultIDs, 'top1_error', 'global', ValidationTop1Error)
* Concatenate the values for the plot.
get_dl_classifier_param (DLClassifierHandle, 'learning_rate', LearningRate)
TrainingErrors := [TrainingErrors,TrainingTop1Error]
ValidationErrors := [ValidationErrors,ValidationTop1Error]
LearningRates := [LearningRates,LearningRate]
Epochs := [Epochs,PlottedIterations[|Epochs|] / real(NumBatchesInEpoch)]
* Plot validation and error against the epochs in order to
* observe the progress of the training.
plot_dl_classifier_training_progress (TrainingErrors, ValidationErrors, LearningRates, Epochs, NumEpochs, WindowHandle)
if (ValidationTop1Error <= MinValidationError)
write_dl_classifier (DLClassifierHandle, FileName)
MinValidationError := ValidationTop1Error
endif
endif
endfor
* Reduce the learning rate every nth epoch.
if ((Epoch + 1) % LearningRateStepEveryNthEpoch == 0)
set_dl_classifier_param (DLClassifierHandle, 'learning_rate', LearningRate * LearningRateStepRatio)
get_dl_classifier_param (DLClassifierHandle, 'learning_rate', LearningRate)
endif
endfor
stop ()
return ()
(训练结果,此处黄色曲线验证误差率并不好)
clear_dl_classifier (DLClassifierHandle) //清除网络句柄
read_dl_classifier (FileName, DLClassifierHandle) //读取序列化网络模型
*计算混淆矩
get_error_for_confusion_matrix (ValidationImages, DLClassifierHandle, Top1ClassValidation)
apply_dl_classifier_batchwise (Images, DLClassifierHandle, DLClassifierResultIDsTest, PredictedClasses, Confidences)
*
Top1PredictedClasses := []
for Index := 0 to PredictedClasses.length() - 1 by 1
Top1PredictedClasses := [Top1PredictedClasses,PredictedClasses.at(Index)[0]]
endfor
return ()
gen_confusion_matrix (ValidationLabels, Top1ClassValidation, [], [], WindowHandle, ConfusionMatrix) //生成混淆矩模型
* This procedure computes a confusion matrix.
* Therefore, it compares the classes
* assigned in GroundTruthLabels and PredictedClasses.
* The resulting confusion matrix can be
* visualized, returned, or both.
* In each case, the output can be changed
* via generic parameters using GenParamName and GenParamValue.
* For the visualization, the graphics window
* must be specified with WindowHandle.
*
if (|GroundTruthLabels| != |PredictedClasses|)
throw ('Number of ground truth labels and predicted classes must be equal.')
endif
*
* Set generic parameter defaults.
DisplayMatrix := 'absolute'
ReturnMatrix := 'absolute'
DisplayColor := 'true'
DisplayColumnWidth := 'minimal'
*
* Parse generic parameters.
for GenParamIndex := 0 to |GenParamName| - 1 by 1
if (GenParamName[GenParamIndex] == 'display_matrix')
* Set 'display_matrix'.
DisplayMatrix := GenParamValue[GenParamIndex]
elseif (GenParamName[GenParamIndex] == 'return_matrix')
* Set 'return_matrix'.
ReturnMatrix := GenParamValue[GenParamIndex]
elseif (GenParamName[GenParamIndex] == 'display_color')
* Set 'display_color'.
DisplayColor := GenParamValue[GenParamIndex]
elseif (GenParamName[GenParamIndex] == 'display_column_width')
* Set 'display_column_width'.
DisplayColumnWidth := GenParamValue[GenParamIndex]
else
throw ('Unknown generic parameter: \'' + GenParamName[GenParamIndex] + '\'')
endif
endfor
*
if (DisplayMatrix == 'relative' or ReturnMatrix == 'relative' or DisplayColor == 'true')
CalculateRelativeMatrix := 1
else
CalculateRelativeMatrix := 0
endif
*
* Calculate the confusion matrix with absolute values
* and the confusion matrix with relative errors.
* We start with an empty matrix
* and add the number of matching labels.
Classes := uniq(sort(GroundTruthLabels))
NumClasses := |Classes|
create_matrix (NumClasses, NumClasses, 0, AbsoluteMatrixID)
if (CalculateRelativeMatrix)
create_matrix (NumClasses, NumClasses, 0, RelativeMatrixID)
endif
for ColumnMatrix := 0 to NumClasses - 1 by 1
Class := Classes[ColumnMatrix]
ThisLabel := GroundTruthLabels [==] Class
if (CalculateRelativeMatrix)
* Obtain the number of ground truth labels per class.
NumClassGroundTruth := sum(ThisLabel)
endif
for RowMatrix := 0 to NumClasses - 1 by 1
* Select classes for this row/column.
PredictedClass := Classes[RowMatrix]
* Check whether the input data
* corresponds to these classes.
ThisPredictedClass := PredictedClasses [==] PredictedClass
* Count the number of elements where the predicted class
* matches the ground truth label.
NumMatches := sum((ThisLabel + ThisPredictedClass) [==] 2)
* Set value in matrix.
set_value_matrix (AbsoluteMatrixID, RowMatrix, ColumnMatrix, NumMatches)
if (CalculateRelativeMatrix)
if (NumClassGroundTruth > 0)
RelativeError := real(NumMatches) / NumClassGroundTruth
else
RelativeError := 0
endif
set_value_matrix (RelativeMatrixID, RowMatrix, ColumnMatrix, RelativeError)
endif
endfor
endfor
*
* Return the result.
if (ReturnMatrix == 'absolute')
copy_matrix (AbsoluteMatrixID, ConfusionMatrix)
elseif (ReturnMatrix == 'relative')
copy_matrix (RelativeMatrixID, ConfusionMatrix)
elseif (ReturnMatrix == 'none')
* No matrix is returned.
else
throw ('Unsupported mode for \'return_matrix\'')
endif
*
* Display the matrix.
if (DisplayMatrix != 'none')
*
* Find maximal string width and set display position parameters.
StringWidths := []
* Get the string width of each class.
for StringIndex := 0 to |Classes| - 1 by 1
String := Classes[StringIndex]
get_string_extents (WindowHandle, String, Ascent, Descent, StringWidth, StringHeight)
StringWidths := [StringWidths,StringWidth]
endfor
* The columns should have a minimum width for 4 characters.
get_string_extents (WindowHandle, 'test', Ascent, Descent, StringWidth, StringHeight)
MaxStringWidth := max2(max(StringWidths),StringWidth)
* Get the maximum string width
* and resize the window accordingly.
RowStart := 80
RowDistance := StringHeight + 10
RowEnd := StringHeight * 7
ColumnStart := 50 + MaxStringWidth
ColumnOffset := 20
ColumnEnd := ColumnOffset
*
* Adapt the window size to fit the confusion matrix.
if (DisplayColumnWidth == 'minimal')
* Every column of the confusion matrix is as narrow as possible
* based to the respective string widths.
Width := sum(StringWidths) + ColumnOffset * NumClasses + ColumnStart + ColumnEnd
elseif (DisplayColumnWidth == 'equal')
* Every column of the confusion matrix should have the same width.
* based on the maximum string width.
Width := (MaxStringWidth + ColumnOffset) * NumClasses + ColumnStart + ColumnEnd
else
throw ('')
endif
Height := RowDistance * NumClasses + RowStart + RowEnd
dev_set_window (WindowHandle)
dev_clear_window ()
*
* Set reasonable limits for graphics window (adapt if necessary).
WidthLimit := [450,1920]
HeightLimit := [250,1080]
if (Width > WidthLimit[1] or Height > HeightLimit[1])
throw ('Confusion Matrix does not fit into graphics window. Please adapt font and/or size limits.')
endif
dev_resize_window_fit_size (0, 0, Width, Height, WidthLimit, HeightLimit)
*
* Get display coordinates.
* Get row coordinates for display.
TextRow := []
for ColumnMatrix := 0 to NumClasses - 1 by 1
TextRow := [TextRow,[0:RowDistance:(NumClasses - 1) * RowDistance]]
endfor
* Get column coordinates for display.
TextColumn := []
for Index := 0 to NumClasses - 1 by 1
TextColumn := [TextColumn,gen_tuple_const(NumClasses,ColumnStart)]
if (DisplayColumnWidth == 'minimal')
ColumnStart := ColumnStart + StringWidths[Index] + ColumnOffset
elseif (DisplayColumnWidth == 'equal')
ColumnStart := ColumnStart + MaxStringWidth + ColumnOffset
endif
endfor
* Display the confusion matrix with a margin from the top.
TextRow := TextRow + RowStart
* Display axis titles.
dev_disp_text ('Ground truth labels', 'window', 'top', 'right', 'white', 'box', 'false')
dev_disp_text ('Predicted classes', 'window', 'bottom', 'left', 'white', 'box', 'false')
for Index := 0 to |Classes| - 1 by 1
Text := Classes[Index]
* Display predicted class names.
Row := TextRow[Index]
Column := TextColumn[0] - MaxStringWidth - ColumnOffset
dev_disp_text (Text, 'window', Row, Column, 'light gray', 'box', 'false')
* Display ground truth label names.
Row := TextRow[0] - RowDistance
Column := TextColumn[Index * NumClasses]
dev_disp_text (Text, 'window', Row, Column, 'light gray', 'box', 'false')
endfor
*
* Get the confusion matrix values for display.
if (DisplayMatrix == 'absolute')
* Displayed matrix corresponds to the transposed returned matrix.
transpose_matrix (AbsoluteMatrixID, AbsoluteTransposedMatrixID)
get_full_matrix (AbsoluteTransposedMatrixID, MatrixText)
clear_matrix (AbsoluteTransposedMatrixID)
* Align the numbers right.
max_matrix (AbsoluteMatrixID, 'full', MatrixMaxID)
get_full_matrix (MatrixMaxID, MaxValue)
clear_matrix (MatrixMaxID)
StringConversion := int(ceil(log10(MaxValue))) + '.0f'
MatrixText := MatrixText$StringConversion
else
* Displayed matrix corresponds to the transposed returned matrix.
transpose_matrix (RelativeMatrixID, RelativeTransposedMatrixID)
get_full_matrix (RelativeTransposedMatrixID, MatrixText)
clear_matrix (RelativeTransposedMatrixID)
MatrixText := MatrixText$'.2f'
endif
* Set color for displayed confusion matrix.
if (DisplayColor == 'true')
tuple_gen_const (|MatrixText|, '#666666', TextColor)
* Use the relative values to adapt the color of the text.
transpose_matrix (RelativeMatrixID, RelativeTransposedMatrixID)
get_full_matrix (RelativeTransposedMatrixID, RelativeValues)
clear_matrix (RelativeTransposedMatrixID)
* Set the colors and respective thresholds for the off-diagonal values.
Thresholds := [0.0,0.05,0.1,0.2]
Colors := ['#8C4D4D','#B33333','#D91A1A','#FF0000']
for Index := 0 to |Thresholds| - 1 by 1
tuple_greater_elem (RelativeValues, Thresholds[Index], Greater)
tuple_find (Greater, 1, Indices)
if (Indices != -1)
tuple_replace (TextColor, Indices, Colors[Index], TextColor)
else
break
endif
endfor
* Set the colors and respective thresholds for the diagonal values.
Thresholds := [-0.01,0.60,0.80,0.90,0.95,0.98]
Colors := ['#666666','#508650','#419C41','#2BBD2B','#15DE15','#00FF00']
for DiagonalIndex := 0 to NumClasses - 1 by 1
get_value_matrix (RelativeMatrixID, DiagonalIndex, DiagonalIndex, Value)
for Index := 0 to |Thresholds| - 1 by 1
if (Value > Thresholds[Index])
TextColor[DiagonalIndex * (NumClasses + 1)] := Colors[Index]
else
break
endif
endfor
endfor
else
* Default value for the text color.
tuple_gen_const (|MatrixText|, 'white', TextColor)
endif
*
* Display confusion matrix.
dev_disp_text (MatrixText, 'window', TextRow, TextColumn, TextColor, 'box', 'false')
*
* Clean up.
if (CalculateRelativeMatrix)
clear_matrix (RelativeMatrixID)
endif
clear_matrix (AbsoluteMatrixID)
endif
return ()* This procedure computes a confusion matrix.
* Therefore, it compares the classes
* assigned in GroundTruthLabels and PredictedClasses.
* The resulting confusion matrix can be
* visualized, returned, or both.
* In each case, the output can be changed
* via generic parameters using GenParamName and GenParamValue.
* For the visualization, the graphics window
* must be specified with WindowHandle.
*
if (|GroundTruthLabels| != |PredictedClasses|)
throw ('Number of ground truth labels and predicted classes must be equal.')
endif
*
* Set generic parameter defaults.
DisplayMatrix := 'absolute'
ReturnMatrix := 'absolute'
DisplayColor := 'true'
DisplayColumnWidth := 'minimal'
*
* Parse generic parameters.
for GenParamIndex := 0 to |GenParamName| - 1 by 1
if (GenParamName[GenParamIndex] == 'display_matrix')
* Set 'display_matrix'.
DisplayMatrix := GenParamValue[GenParamIndex]
elseif (GenParamName[GenParamIndex] == 'return_matrix')
* Set 'return_matrix'.
ReturnMatrix := GenParamValue[GenParamIndex]
elseif (GenParamName[GenParamIndex] == 'display_color')
* Set 'display_color'.
DisplayColor := GenParamValue[GenParamIndex]
elseif (GenParamName[GenParamIndex] == 'display_column_width')
* Set 'display_column_width'.
DisplayColumnWidth := GenParamValue[GenParamIndex]
else
throw ('Unknown generic parameter: \'' + GenParamName[GenParamIndex] + '\'')
endif
endfor
*
if (DisplayMatrix == 'relative' or ReturnMatrix == 'relative' or DisplayColor == 'true')
CalculateRelativeMatrix := 1
else
CalculateRelativeMatrix := 0
endif
*
* Calculate the confusion matrix with absolute values
* and the confusion matrix with relative errors.
* We start with an empty matrix
* and add the number of matching labels.
Classes := uniq(sort(GroundTruthLabels))
NumClasses := |Classes|
create_matrix (NumClasses, NumClasses, 0, AbsoluteMatrixID)
if (CalculateRelativeMatrix)
create_matrix (NumClasses, NumClasses, 0, RelativeMatrixID)
endif
for ColumnMatrix := 0 to NumClasses - 1 by 1
Class := Classes[ColumnMatrix]
ThisLabel := GroundTruthLabels [==] Class
if (CalculateRelativeMatrix)
* Obtain the number of ground truth labels per class.
NumClassGroundTruth := sum(ThisLabel)
endif
for RowMatrix := 0 to NumClasses - 1 by 1
* Select classes for this row/column.
PredictedClass := Classes[RowMatrix]
* Check whether the input data
* corresponds to these classes.
ThisPredictedClass := PredictedClasses [==] PredictedClass
* Count the number of elements where the predicted class
* matches the ground truth label.
NumMatches := sum((ThisLabel + ThisPredictedClass) [==] 2)
* Set value in matrix.
set_value_matrix (AbsoluteMatrixID, RowMatrix, ColumnMatrix, NumMatches)
if (CalculateRelativeMatrix)
if (NumClassGroundTruth > 0)
RelativeError := real(NumMatches) / NumClassGroundTruth
else
RelativeError := 0
endif
set_value_matrix (RelativeMatrixID, RowMatrix, ColumnMatrix, RelativeError)
endif
endfor
endfor
*
* Return the result.
if (ReturnMatrix == 'absolute')
copy_matrix (AbsoluteMatrixID, ConfusionMatrix)
elseif (ReturnMatrix == 'relative')
copy_matrix (RelativeMatrixID, ConfusionMatrix)
elseif (ReturnMatrix == 'none')
* No matrix is returned.
else
throw ('Unsupported mode for \'return_matrix\'')
endif
*
* Display the matrix.
if (DisplayMatrix != 'none')
*
* Find maximal string width and set display position parameters.
StringWidths := []
* Get the string width of each class.
for StringIndex := 0 to |Classes| - 1 by 1
String := Classes[StringIndex]
get_string_extents (WindowHandle, String, Ascent, Descent, StringWidth, StringHeight)
StringWidths := [StringWidths,StringWidth]
endfor
* The columns should have a minimum width for 4 characters.
get_string_extents (WindowHandle, 'test', Ascent, Descent, StringWidth, StringHeight)
MaxStringWidth := max2(max(StringWidths),StringWidth)
* Get the maximum string width
* and resize the window accordingly.
RowStart := 80
RowDistance := StringHeight + 10
RowEnd := StringHeight * 7
ColumnStart := 50 + MaxStringWidth
ColumnOffset := 20
ColumnEnd := ColumnOffset
*
* Adapt the window size to fit the confusion matrix.
if (DisplayColumnWidth == 'minimal')
* Every column of the confusion matrix is as narrow as possible
* based to the respective string widths.
Width := sum(StringWidths) + ColumnOffset * NumClasses + ColumnStart + ColumnEnd
elseif (DisplayColumnWidth == 'equal')
* Every column of the confusion matrix should have the same width.
* based on the maximum string width.
Width := (MaxStringWidth + ColumnOffset) * NumClasses + ColumnStart + ColumnEnd
else
throw ('')
endif
Height := RowDistance * NumClasses + RowStart + RowEnd
dev_set_window (WindowHandle)
dev_clear_window ()
*
* Set reasonable limits for graphics window (adapt if necessary).
WidthLimit := [450,1920]
HeightLimit := [250,1080]
if (Width > WidthLimit[1] or Height > HeightLimit[1])
throw ('Confusion Matrix does not fit into graphics window. Please adapt font and/or size limits.')
endif
dev_resize_window_fit_size (0, 0, Width, Height, WidthLimit, HeightLimit)
*
* Get display coordinates.
* Get row coordinates for display.
TextRow := []
for ColumnMatrix := 0 to NumClasses - 1 by 1
TextRow := [TextRow,[0:RowDistance:(NumClasses - 1) * RowDistance]]
endfor
* Get column coordinates for display.
TextColumn := []
for Index := 0 to NumClasses - 1 by 1
TextColumn := [TextColumn,gen_tuple_const(NumClasses,ColumnStart)]
if (DisplayColumnWidth == 'minimal')
ColumnStart := ColumnStart + StringWidths[Index] + ColumnOffset
elseif (DisplayColumnWidth == 'equal')
ColumnStart := ColumnStart + MaxStringWidth + ColumnOffset
endif
endfor
* Display the confusion matrix with a margin from the top.
TextRow := TextRow + RowStart
* Display axis titles.
dev_disp_text ('Ground truth labels', 'window', 'top', 'right', 'white', 'box', 'false')
dev_disp_text ('Predicted classes', 'window', 'bottom', 'left', 'white', 'box', 'false')
for Index := 0 to |Classes| - 1 by 1
Text := Classes[Index]
* Display predicted class names.
Row := TextRow[Index]
Column := TextColumn[0] - MaxStringWidth - ColumnOffset
dev_disp_text (Text, 'window', Row, Column, 'light gray', 'box', 'false')
* Display ground truth label names.
Row := TextRow[0] - RowDistance
Column := TextColumn[Index * NumClasses]
dev_disp_text (Text, 'window', Row, Column, 'light gray', 'box', 'false')
endfor
*
* Get the confusion matrix values for display.
if (DisplayMatrix == 'absolute')
* Displayed matrix corresponds to the transposed returned matrix.
transpose_matrix (AbsoluteMatrixID, AbsoluteTransposedMatrixID)
get_full_matrix (AbsoluteTransposedMatrixID, MatrixText)
clear_matrix (AbsoluteTransposedMatrixID)
* Align the numbers right.
max_matrix (AbsoluteMatrixID, 'full', MatrixMaxID)
get_full_matrix (MatrixMaxID, MaxValue)
clear_matrix (MatrixMaxID)
StringConversion := int(ceil(log10(MaxValue))) + '.0f'
MatrixText := MatrixText$StringConversion
else
* Displayed matrix corresponds to the transposed returned matrix.
transpose_matrix (RelativeMatrixID, RelativeTransposedMatrixID)
get_full_matrix (RelativeTransposedMatrixID, MatrixText)
clear_matrix (RelativeTransposedMatrixID)
MatrixText := MatrixText$'.2f'
endif
* Set color for displayed confusion matrix.
if (DisplayColor == 'true')
tuple_gen_const (|MatrixText|, '#666666', TextColor)
* Use the relative values to adapt the color of the text.
transpose_matrix (RelativeMatrixID, RelativeTransposedMatrixID)
get_full_matrix (RelativeTransposedMatrixID, RelativeValues)
clear_matrix (RelativeTransposedMatrixID)
* Set the colors and respective thresholds for the off-diagonal values.
Thresholds := [0.0,0.05,0.1,0.2]
Colors := ['#8C4D4D','#B33333','#D91A1A','#FF0000']
for Index := 0 to |Thresholds| - 1 by 1
tuple_greater_elem (RelativeValues, Thresholds[Index], Greater)
tuple_find (Greater, 1, Indices)
if (Indices != -1)
tuple_replace (TextColor, Indices, Colors[Index], TextColor)
else
break
endif
endfor
* Set the colors and respective thresholds for the diagonal values.
Thresholds := [-0.01,0.60,0.80,0.90,0.95,0.98]
Colors := ['#666666','#508650','#419C41','#2BBD2B','#15DE15','#00FF00']
for DiagonalIndex := 0 to NumClasses - 1 by 1
get_value_matrix (RelativeMatrixID, DiagonalIndex, DiagonalIndex, Value)
for Index := 0 to |Thresholds| - 1 by 1
if (Value > Thresholds[Index])
TextColor[DiagonalIndex * (NumClasses + 1)] := Colors[Index]
else
break
endif
endfor
endfor
else
* Default value for the text color.
tuple_gen_const (|MatrixText|, 'white', TextColor)
endif
*
* Display confusion matrix.
dev_disp_text (MatrixText, 'window', TextRow, TextColumn, TextColor, 'box', 'false')
*
* Clean up.
if (CalculateRelativeMatrix)
clear_matrix (RelativeMatrixID)
endif
clear_matrix (AbsoluteMatrixID)
endif
return ()
(混淆矩阵)
dev_disp_text ('Validation data', 'window', 'top', 'left', 'gray', 'box', 'false')
dev_disp_text ('Press Run (F5) to continue', 'window', 'bottom', 'right', 'black', [], [])
stop ()
clear_matrix (ConfusionMatrix)
dev_clear_window ()
clear_dl_classifier (DLClassifierHandle)
read_dl_classifier (FileName, DLClassifierHandle)
set_dl_classifier_param (DLClassifierHandle, 'batch_size', 1)
set_dl_classifier_param (DLClassifierHandle, 'runtime_init', 'immediately')
dev_resize_window_fit_size (0, 0, WindowWidth, WindowHeight, -1, -1)
ResultDir := 'result/'
file_exists(ResultDir, FileExists)
if(FileExists)
remove_dir_recursively (ResultDir) //将已存在的result删除
*remove_dir_recursively (ResultDir + 'labels/')
endif
make_dir (ResultDir) //生成目录
*使用训练好的模型进行检测
set_display_font (WindowHandle, 30, 'mono', 'true', 'false')
list_files ('Test', 'files', Files)
for Index := 0 to |Files|-1 by 1
read_image (Image, Files[Index])
zoom_image_size (Image, Image, DlImageWidth, DlImageHeight, 'constant')
convert_image_type (Image, Image, 'real')
RescaleRange:=(DlRangeMax - DlRangeMin)/255.0
scale_image (Image, Image, RescaleRange, DlRangeMin)
apply_dl_classifier (Image, DLClassifierHandle, DLClassifierResultHandle)
get_dl_classifier_result (DLClassifierResultHandle, 'all', 'predicted_classes', PredictedClass)
*clear_dl_1classifier_result (DLClassifierResultHandle)
*
dev_display (Image)
Text := 'Predicted class: ' + PredictedClass
if (PredictedClass == 'OK')
disp_message (WindowHandle, Text, 'window', 12, 12, 'green', 'false')
parse_filename (ImageFiles[Index], BaseName, Extension, Directory)
dump_window (WindowHandle, 'png', ResultDir + BaseName)
else
disp_message (WindowHandle, Text, 'window', 12, 12, 'red', 'false')
parse_filename (ImageFiles[Index], BaseName, Extension, Directory)
dump_window (WindowHandle, 'png', ResultDir + BaseName)
endif
stop ()
endfor
clear_dl_classifier (DLClassifierHandle)
(模型还需要优化,小女孩识别应该是:OK,这里误判为NG,10张错一张)
优化思路:
1.增加小女孩这样的训练集 2.增大数据集、增加训练epoch、适当修改学习率变化参数
第二次训练:
(识别错误图)10张错1张