Halcon20.11深度学习语义分割模型

豆浩宇

于 2024-08-27 13:48:43 发布

阅读量942

点赞数 22

文章标签：深度学习目标检测人工智能 c++ 计算机视觉算法机器学习

本文链接：https://blog.csdn.net/Douhaoyu/article/details/141600914

版权

1.前言：深度目标检测模型已经可以满足一大部分的检测需求，但是在逐像素精度分割方向是无法做到的。这时候就需要训练深度语义分割模型，标注工具依然使用的MVTec Deep Learning Tool 24.05。实现顺序依然是先用标注工具把所有的图标注完毕后，导出标注数据集，即可利用此数据集，在halcon代码中训练。
在这里插入图片描述

2.上干货，深度语义分割训练源码：
模型训练预处理准备阶段********************
dev_update_off ()
*

*** Set Input/Output paths. ***

TotalPath:=‘E:/HalconDL_Project/瓶盖划伤语义分割/标注/’

Directory with image data.
ImageDir := TotalPath+‘Images’
Directory with ground truth segmentation images.
SegmentationDir := TotalPath+‘Bottle_labels’
All example data is written to this folder.
ExampleDataDir := TotalPath + ‘segment_Bottle_defects_data’
Dataset directory basename for any outputs written by preprocess_dl_dataset.
This name will be extended by the image dimensions the dataset will have after preprocessing.
DataDirectoryBaseName := ExampleDataDir + ‘/dldataset_Bottle_’
Store preprocess params separately in order to use it e.g. during inference.
PreprocessParamFileBaseName := ‘/dl_preprocess_param.hdict’
*标注数据集路径
DirectPath_Bottle:=TotalPath + ‘Bottle.hdict’

*** Set parameters ***

Class names.
ClassNames := [‘Bacground’,‘Dirty’,‘HuaShang’]
Class IDs.
ClassIDs := [0,1,2]
Percentages for splitting the dataset.
TrainingPercent := 70
ValidationPercent := 15
Image dimensions the images are rescaled to during preprocessing.
ImageWidth := 400
ImageHeight := 400
ImageNumChannels := 3
Gray value range for gray value normalization of the images.
ImageRangeMin := -127
ImageRangeMax := 128
Further parameters for image preprocessing.
NormalizationType := ‘none’
DomainHandling := ‘full_domain’
IgnoreClassIDs := []
SetBackgroundID := []
ClassIDsBackground := []
In order to get a reproducible split we set a random seed.
This means that re-running the script results in the same split of DLDataset.
SeedRand := 42

** Read the labeled data and split it into train/validation and test ***

Set the random seed.
set_system (‘seed_rand’, SeedRand)
Read the dataset.
read_dict (DirectPath_Bottle, [], [], DLDataset)
read_dl_dataset_segmentation (ImageDir, SegmentationDir, ClassNames, ClassIDs, [], [], [], DLDataset)
Generate the split.
split_dl_dataset (DLDataset, TrainingPercent, ValidationPercent, [])

** Preprocess the dataset ***

Create the output directory if it does not exist yet.
file_exists (ExampleDataDir, FileExists)
if (not FileExists)
make_dir (ExampleDataDir)
endif
Create preprocess param.
create_dl_preprocess_param (‘segmentation’, ImageWidth, ImageHeight, ImageNumChannels, ImageRangeMin, ImageRangeMax, NormalizationType, DomainHandling, IgnoreClassIDs, SetBackgroundID, ClassIDsBackground, [], DLPreprocessParam)
Dataset directory for any outputs written by preprocess_dl_dataset.
DataDirectory := DataDirectoryBaseName + ImageWidth + ‘x’ + ImageHeight
Preprocess the dataset. This might take a few minutes.
create_dict (GenParam)
set_dict_tuple (GenParam, ‘overwrite_files’, true)
preprocess_dl_dataset (DLDataset, DataDirectory, DLPreprocessParam, GenParam, DLDatasetFilename)
Store preprocess params separately in order to use it e.g. during inference.
PreprocessParamFile := DataDirectory + PreprocessParamFileBaseName
write_dict (DLPreprocessParam, PreprocessParamFile, [], [])

** Preview the preprocessed dataset ***

Before moving on to training, it is recommended to check the preprocessed dataset.
Display the DLSamples for 10 randomly selected train images.
get_dict_tuple (DLDataset, ‘samples’, DatasetSamples)
find_dl_samples (DatasetSamples, ‘split’, ‘train’, ‘match’, SampleIndices)
tuple_shuffle (SampleIndices, ShuffledIndices)
read_dl_samples (DLDataset, ShuffledIndices[0:9], DLSampleBatchDisplay)

create_dict (WindowHandleDict)
for Index := 0 to |DLSampleBatchDisplay| - 1 by 1
* Loop over samples in DLSampleBatchDisplay.
dev_display_dl_data (DLSampleBatchDisplay[Index], [], DLDataset, [‘image’,‘segmentation_image_ground_truth’], [], WindowHandleDict)
get_dict_tuple (WindowHandleDict, ‘segmentation_image_ground_truth’, WindowHandleImage)
dev_set_window (WindowHandleImage[1])
Text := ‘Press Run (F5) to continue’
dev_disp_text (Text, ‘window’, 400, 40, ‘black’, [], [])
stop ()
endfor
*

Close windows that have been used for visualization.
dev_close_window_dict (WindowHandleDict)

模型训练阶段*****************
dev_update_off ()

Training can be performed on a GPU or CPU.
See the respective system requirements in the Installation Guide.
If possible a GPU is used in this example.
In case you explicitely wish to run this example on the CPU,
choose the CPU device instead.
query_available_dl_devices ([‘runtime’,‘runtime’], [‘gpu’,‘cpu’], DLDeviceHandles)
if (|DLDeviceHandles| == 0)
throw (‘No supported device found to continue this example.’)
endif
Due to the filter used in query_available_dl_devices, the first device is a GPU, if available.
DLDevice := DLDeviceHandles[0]
get_dl_device_param (DLDevice, ‘type’, DLDeviceType)
if (DLDeviceType == ‘cpu’)
- The number of used threads may have an impact
- on the training duration.
  NumThreadsTraining := 4
  set_system (‘thread_num’, NumThreadsTraining)
  endif

*** Set Input/Output paths. ***

All example data is written to this folder.
TotalPath:=‘E:/HalconDL_Project/瓶盖划伤语义分割/标注/’
ExampleDataDir := TotalPath + ‘segment_Bottle_defects_data’
File path of the preprocessed DLDataset.
Note: Adapt DataDirectory after preprocessing with another image size.
DataDirectory := ExampleDataDir + ‘/dldataset_Bottle_400x400’
DLDatasetFileName := DataDirectory + ‘/dl_dataset.hdict’
Output path for the final trained model.
FinalModelBaseName := ExampleDataDir + ‘/final_dl_model_segmentation’
Output path of the best evaluated model.
BestModelBaseName := ExampleDataDir + ‘/best_dl_model_segmentation’

*** Set basic parameters ***

The following parameters need to be adapted frequently.
Model parameters.
The segmentation model to be retrained.
ModelFileName := ‘pretrained_dl_segmentation_enhanced.hdl’
Batch size.
If set to ‘maximum’, the batch size is set by set_dl_model_param_max_gpu_batch_size
if the runtime ‘gpu’ is given.
BatchSize := ‘maximum’
Initial learning rate.
InitialLearningRate := 0.0001
Momentum should be high if batch size is small.
Momentum := 0.99
Parameters used by train_dl_model.
Number of epochs to train the model.
NumEpochs := 10
Evaluation interval (in epochs) to calculate evaluation measures on validation split.
EvaluationIntervalEpochs := 1
Change the learning rate in the following epochs, e.g. [15, 30].
Set it to [] if the learning rate should not be changed.
ChangeLearningRateEpochs := []
Change the learning rate to the following values, e.g. InitialLearningRate * [0.1, 0.01].
The tuple has to be of the same length as ChangeLearningRateEpochs.
ChangeLearningRateValues := []

*** Set advanced parameters. ***

The following parameters might need to be changed in rare cases.
Model parameter.
Use [] for default weight prior.
WeightPrior := []
Parameters of train_dl_model.
Control whether training progress is displayed (true/false).
EnableDisplay := true
Set a random seed for training.
RandomSeed := 42
set_system (‘seed_rand’, RandomSeed)
In order to obtain nearly deterministic training results on the same GPU
(system, driver, cuda-version) you could specify “cudnn_deterministic” as
“true”. Note, that this could slow down training a bit.
set_system (‘cudnn_deterministic’, ‘true’)
Set generic parameters of create_dl_train_param.
Please see the documentation of create_dl_train_param for an overview on all available parameters.
GenParamName := []
GenParamValue := []
Change strategies.
It is possible to change model parameters during training.
Here, we change the learning rate if specified above.
if (|ChangeLearningRateEpochs| > 0)
create_dict (ChangeStrategy)
- Specify the model parameter to be changed.
  set_dict_tuple (ChangeStrategy, ‘model_param’, ‘learning_rate’)
- Start the parameter value at ‘initial_value’.
  set_dict_tuple (ChangeStrategy, ‘initial_value’, InitialLearningRate)
- Change the parameter value at each ‘epochs’ step.
  set_dict_tuple (ChangeStrategy, ‘epochs’, ChangeLearningRateEpochs)
- Change the parameter value to the corresponding value in values.
  set_dict_tuple (ChangeStrategy, ‘values’, ChangeLearningRateValues)
- Collect all change strategies as input.
  GenParamName := [GenParamName,‘change’]
  GenParamValue := [GenParamValue,ChangeStrategy]
  endif
Serialization strategies.
There are several options for saving intermediate models to disk (see create_dl_train_param).
Here, the best and final model are saved to the paths set above.
create_dict (SerializationStrategy)
set_dict_tuple (SerializationStrategy, ‘type’, ‘best’)
set_dict_tuple (SerializationStrategy, ‘basename’, BestModelBaseName)
GenParamName := [GenParamName,‘serialize’]
GenParamValue := [GenParamValue,SerializationStrategy]
create_dict (SerializationStrategy)
set_dict_tuple (SerializationStrategy, ‘type’, ‘final’)
set_dict_tuple (SerializationStrategy, ‘basename’, FinalModelBaseName)
GenParamName := [GenParamName,‘serialize’]
GenParamValue := [GenParamValue,SerializationStrategy]
Display parameters.
In this example, the evaluation measure for the training spit is not displayed during
training (default). If you want to do so, select a certain percentage of the training
samples used to evaluate the model during training. A lower percentage helps to speed
up the evaluation.
SelectedPercentageTrainSamples := 0
Set the x-axis argument of the training plots.
XAxisLabel := ‘epochs’
create_dict (DisplayParam)
set_dict_tuple (DisplayParam, ‘selected_percentage_train_samples’, SelectedPercentageTrainSamples)
set_dict_tuple (DisplayParam, ‘x_axis_label’, XAxisLabel)
GenParamName := [GenParamName,‘display’]
GenParamValue := [GenParamValue,DisplayParam]

*** Read model and dataset. ***

Check if all necessary files exist.
check_data_availability (ExampleDataDir, DLDatasetFileName)
Read the preprocessed DLDataset file.
read_dict (DLDatasetFileName, [], [], DLDataset)
Read in the model that was initialized during preprocessing.
read_dl_model (ModelFileName, DLModelHandle)

set_dl_model_param (DLModelHandle, ‘device’, DLDevice)
*

*** Set model parameters. ***

Set model parameters according to preprocessing parameters.
get_dict_tuple (DLDataset, ‘preprocess_param’, DLPreprocessParam)
get_dict_tuple (DLDataset, ‘class_ids’, ClassIDs)
set_dl_model_param_based_on_preprocessing (DLModelHandle, DLPreprocessParam, ClassIDs)
Set model hyper-parameters as specified above.
set_dl_model_param (DLModelHandle, ‘learning_rate’, InitialLearningRate)
set_dl_model_param (DLModelHandle, ‘momentum’, Momentum)
if (BatchSize == ‘maximum’ and DLDeviceType == ‘gpu’)
set_dl_model_param_max_gpu_batch_size (DLModelHandle, 100)
else
if (BatchSize == ‘maximum’ and DLDeviceType == ‘cpu’)
* Please set a suitable batch size in case of ‘cpu’
* training before continuing.
stop ()
endif
set_dl_model_param (DLModelHandle, ‘batch_size’, 1)
endif
if (|WeightPrior| > 0)
set_dl_model_param (DLModelHandle, ‘weight_prior’, WeightPrior)
endif
set_dl_model_param (DLModelHandle, ‘runtime_init’, ‘immediately’)

*** Train the model. ***

Create the generic train parameter dictionary.
create_dl_train_param (DLModelHandle, NumEpochs, EvaluationIntervalEpochs, EnableDisplay, RandomSeed, GenParamName, GenParamValue, TrainParam)
Start the training by calling the training operator
train_dl_model_batch () within the following procedure.
train_dl_model (DLDataset, DLModelHandle, TrainParam, 0.0, TrainResults, TrainInfos, EvaluationInfos)
Stop after the training has finished, before closing the windows.
dev_disp_text (‘Press Run (F5) to continue’, ‘window’, ‘bottom’, ‘right’, ‘black’, [], [])
stop ()
Close training windows.
dev_close_window ()
dev_close_window ()

模型评估阶段************************
dev_update_off ()
*

By default, this example uses a model pretrained by MVTec. To use the model
which was trained in part 2 of this example series, set the following
variable to false.
UsePretrainedModel := true
Evaluation can be performed on a GPU or CPU.
See the respective system requirements in the Installation Guide.
If possible a GPU is used in this example.
In case you explicitely wish to run this example on the CPU,
choose the CPU device instead.
query_available_dl_devices ([‘runtime’,‘runtime’], [‘gpu’,‘cpu’], DLDeviceHandles)
if (|DLDeviceHandles| == 0)
throw (‘No supported device found to continue this example.’)
endif
Due to the filter used in query_available_dl_devices, the first device is a GPU, if available.
DLDevice := DLDeviceHandles[0]

** Set paths and parameters for the evaluation ***

Paths.
TotalPath:=‘E:/HalconDL_Project/瓶盖划伤语义分割/标注/’
Project directory for any outputs written by HALCON.
ExampleDataDir := TotalPath + ‘segment_Bottle_defects_data’
File path of the preprocessed DLDataset.
Note: Adapt DataDirectory after preprocessing with another image size.
DataDirectory := ExampleDataDir + ‘/dldataset_Bottle_400x400’
DLDatasetFileName := DataDirectory + ‘/dl_dataset.hdict’
Path of the retrained segmentation model.
RetrainedModelFileName := ExampleDataDir + ‘/best_dl_model_segmentation.hdl’
Evaluation parameters.
Evaluation measures.
SegmentationMeasures := [‘mean_iou’,‘pixel_accuracy’,‘class_pixel_accuracy’,‘pixel_confusion_matrix’]
Batch size used during evaluation.
BatchSize := 1
Display some segmentation results after determining the best model.
NumDisplay := 3

** Evaluation of the model ***

Check if all necessary files exist.
check_data_availability_COPY_1 (ExampleDataDir, DLDatasetFileName, RetrainedModelFileName, UsePretrainedModel)
Read the retrained model.
read_dl_model (RetrainedModelFileName, DLModelHandle)

set_dl_model_param (DLModelHandle, ‘batch_size’, BatchSize)
*
set_dl_model_param (DLModelHandle, ‘device’, DLDevice)
*

Read the preprocessed DLDataset file.
read_dict (DLDatasetFileName, [], [], DLDataset)
Set parameters for evaluation.
create_dict (GenParamEval)
set_dict_tuple (GenParamEval, ‘measures’, SegmentationMeasures)
set_dict_tuple (GenParamEval, ‘show_progress’, ‘true’)
Evaluate the retrained model.
evaluate_dl_model (DLDataset, DLModelHandle, ‘split’, ‘test’, GenParamEval, EvaluationResult, EvalParams)

** Display the results ***

Display measures.
create_dict (WindowHandleDict)
create_dict (GenParamEvalDisplay)
set_dict_tuple (GenParamEvalDisplay, ‘display_mode’, [‘measures’,‘absolute_confusion_matrix’])
dev_display_segmentation_evaluation (EvaluationResult, EvalParams, GenParamEvalDisplay, WindowHandleDict)

dev_disp_text (‘Press Run (F5) to continue’, ‘window’, ‘bottom’, ‘right’, ‘black’, ‘box’, ‘true’)
stop ()
*

Close window handles.
dev_close_window_dict (WindowHandleDict)

** Visual inspection of images ***

Evaluate the performance of the model qualitatively
by visual inspection of images.
Select test images randomly.
get_dict_tuple (DLDataset, ‘samples’, DatasetSamples)
find_dl_samples (DatasetSamples, ‘split’, ‘test’, ‘match’, SampleIndices)
tuple_shuffle (SampleIndices, ShuffledIndices)
Read the selected samples.
read_dl_samples (DLDataset, ShuffledIndices[0:NumDisplay - 1], DLSampleBatch)
Set parameters for visualization of sample images.
create_dict (WindowHandleDict)
create_dict (GenParamDisplay)
set_dict_tuple (GenParamDisplay, ‘segmentation_exclude_class_ids’, 0)
set_dict_tuple (GenParamDisplay, ‘segmentation_transparency’, ‘80’)
Set batch size of the model to 1.
set_dl_model_param (DLModelHandle, ‘batch_size’, 1)
Apply the retrained model and visualize the results.
for SampleIndex := 0 to NumDisplay - 1 by 1
*
- Apply the model.
  apply_dl_model (DLModelHandle, DLSampleBatch[SampleIndex], [], DLResult)
- Display the result.
  dev_display_dl_data (DLSampleBatch[SampleIndex], DLResult, DLDataset, [‘segmentation_image_ground_truth’,‘segmentation_image_result’], GenParamDisplay, WindowHandleDict)
dev_display_continue_message (WindowHandleDict)
stop ()
endfor
Close the windows.
dev_close_window_dict (WindowHandleDict)
Optimize the memory consumption.
set_dl_model_param (DLModelHandle, ‘optimize_for_inference’, ‘true’)
write_dl_model (DLModelHandle, RetrainedModelFileName)
Close the windows.
dev_close_window_dict (WindowHandleDict)

模型测试*****************************
dev_update_off ()
*

By default, this example uses a model pretrained by MVTec. To use the model
which was trained in part 2 of this example series, set the following
variable to false.
UsePretrainedModel := true
Inference can be done on a GPU or CPU.
See the respective system requirements in the Installation Guide.
If possible a GPU is used in this example.
In case you explicitely wish to run this example on the CPU,
choose the CPU device instead.
query_available_dl_devices ([‘runtime’,‘runtime’], [‘gpu’,‘cpu’], DLDeviceHandles)
if (|DLDeviceHandles| == 0)
throw (‘No supported device found to continue this example.’)
endif
Due to the filter used in query_available_dl_devices, the first device is a GPU, if available.
DLDevice := DLDeviceHandles[0]

** Set paths and parameters for inference ***

We will demonstrate the inference on the example images.
In a real application newly incoming images (not used for training or evaluation)
would be used here.
In this example, we read the images from file.
Directory name with the images to be segmented.
TotalPath:=‘E:/HalconDL_Project/瓶盖划伤语义分割/标注/’
Example data folder containing the outputs of the previous example series.
ExampleDataDir := TotalPath + ‘segment_Bottle_defects_data’
File name of dict containing parameters used for preprocessing.
Note: Adapt DataDirectory after preprocessing with another image size.
DataDirectory := ExampleDataDir + ‘/dldataset_Bottle_400x400’
PreprocessParamFileName := DataDirectory + ‘/dl_preprocess_param.hdict’
Path of the retrained segmentation model.
RetrainedModelFileName := ExampleDataDir + ‘/best_dl_model_segmentation.hdl’
Provide the class names and IDs.
Class names.
ClassNames := [‘BackGround’,‘Dirty’,‘HuaShang’]
Respective class IDs.
ClassIDs := [0,1,2]
Batch Size used during inference.
BatchSizeInference := 1

** Inference ***

Check if all necessary files exist.
check_data_availability_COPY_2 (ExampleDataDir, PreprocessParamFileName, RetrainedModelFileName, UsePretrainedModel)
Read in the retrained model.
read_dl_model (RetrainedModelFileName, DLModelHandle)
Set the batch size.
set_dl_model_param (DLModelHandle, ‘batch_size’, BatchSizeInference)
Initialize the model for inference.
set_dl_model_param (DLModelHandle, ‘device’, DLDevice)
Get the parameters used for preprocessing.
read_dict (PreprocessParamFileName, [], [], DLPreprocessParam)
Set parameters for visualization of results.
create_dict (WindowHandleDict)
create_dict (DatasetInfo)
set_dict_tuple (DatasetInfo, ‘class_ids’, ClassIDs)
set_dict_tuple (DatasetInfo, ‘class_names’, ClassNames)
create_dict (GenParamDisplay)
set_dict_tuple (GenParamDisplay, ‘segmentation_exclude_class_ids’, 0)
set_dict_tuple (GenParamDisplay, ‘segmentation_transparency’, ‘80’)
set_dict_tuple (GenParamDisplay, ‘font_size’, 16)
Image Acquisition 01: Code generated by Image Acquisition 01
list_files (‘E:/HalconDL_Project/瓶盖划伤语义分割/标注/Images’, [‘files’,‘follow_links’], ImageFiles)
tuple_regexp_select (ImageFiles, [‘\.(tif|tiff|gif|bmp|jpg|jpeg|jp2|png|pcx|pgm|ppm|pbm|xwd|ima|hobj)$’,‘ignore_case’], ImageFiles)
for Index := 0 to |ImageFiles| - 1 by 1
read_image (ImageBatch, ImageFiles[Index])
*
- Generate the DLSampleBatch.
  gen_dl_samples_from_images (ImageBatch, DLSampleBatch)
- Preprocess the DLSampleBatch.
  preprocess_dl_samples (DLSampleBatch, DLPreprocessParam)
- Apply the DL model on the DLSampleBatch.
  apply_dl_model (DLModelHandle, DLSampleBatch, [‘segmentation_image’,‘segmentation_confidence’], DLResultBatch)
- Postprocessing and visualization.
- Loop over each sample in the batch.
  for SampleIndex := 0 to BatchSizeInference - 1 by 1
  *
  - Get image.
    get_dict_object (Image, DLSampleBatch[SampleIndex], ‘image’)
  - Get result image.
    get_dict_object (SegmentationImage, DLResultBatch[SampleIndex], ‘segmentation_image’)
  - Postprocessing: Get segmented regions for each class.
    threshold (SegmentationImage, ClassRegions, ClassIDs, ClassIDs)
  - Display results.
    dev_display_dl_data (DLSampleBatch[SampleIndex], DLResultBatch[SampleIndex], DatasetInfo, ‘segmentation_image_result’, GenParamDisplay, WindowHandleDict)
    get_dict_tuple (WindowHandleDict, ‘segmentation_image_result’, WindowHandles)
    dev_set_window (WindowHandles[0])
  - Separate any components of the class regions
  - and display result regions as well as their area.
  - Get area of class regions.
    region_features (ClassRegions, ‘area’, Areas)
  - Here, we do not display the first class, since it is the class ‘good’
  - and we only want to display the defect regions.
    for ClassIndex := 1 to |Areas| - 1 by 1
    if (Areas[ClassIndex] > 0)
    select_obj (ClassRegions, ClassRegion, ClassIndex + 1)
    * Get connected components of the segmented class region.
    connection (ClassRegion, ConnectedRegions)
    area_center (ConnectedRegions, Area, Row, Column)
    for ConnectIndex := 0 to |Area| - 1 by 1
    select_obj (ConnectedRegions, CurrentRegion, ConnectIndex + 1)
    dev_disp_text (ClassNames[ClassIndex] + '\narea: ’ + Area[ConnectIndex] + ‘px’, ‘image’, Row[ConnectIndex] - 10, Column[ConnectIndex] + 10, ‘black’, [], [])
    endfor
    endif
    endfor
  - Display whether the pill is OK, or not.
    dev_display_ok_nok (Areas, WindowHandles[0])
  stop ()
  endfor
  endfor
Close windows.
dev_close_window_dict (WindowHandleDict)