halcon对象检测训练流程

*************************************************halcon对象检测*********************************************
dev_close_window()
dev_get_window(WindowHandle)
dev_set_draw('margin')

**************************************************数据预处理**************************************************
***读取json文件
read_dict ('D:/Source/SourceMe/Halcon深度学习模型/Halcon对象检测模型/Dataset/pill_bag.json', [], [], DictJson)
***获取类别
get_dict_tuple (DictJson, 'categories', CategoryList)
get_dict_param (CategoryList, 'keys', [], CategoryKeys)
***获取图片列表
get_dict_tuple (DictJson, 'images', ImageList)
get_dict_param (ImageList, 'keys', [], ImageKeys)
***获取标注
get_dict_tuple (DictJson, 'annotations', AnnotationList)
get_dict_param (AnnotationList, 'keys', [], AnnotationKeys)

* 设置要检测类
for Index := 0 to |CategoryKeys| - 1 by 1
    try
        get_dict_tuple (CategoryList, CategoryKeys[Index], Category)
        get_dict_tuple (Category, 'id', ID)
        get_dict_tuple (Category, 'name', Name)
    catch (ExceptionCategory)
        throw ('COCO category number ' + Index + ' does not contain the entry id or name.')
    endtry
    ClassIDs[Index] := ID
    ClassNames[Index] := Name
endfor

***创建模型对象检测参数
create_dict(DLModelDetectionParam)
set_dict_tuple(DLModelDetectionParam, 'class_ids', ClassIDs)
set_dict_tuple(DLModelDetectionParam, 'class_names', ClassNames)
set_dict_tuple(DLModelDetectionParam, 'instance_type','rectangle1')
set_dict_tuple(DLModelDetectionParam, 'image_width',512)
set_dict_tuple(DLModelDetectionParam, 'image_height',320)
set_dict_tuple(DLModelDetectionParam, 'max_overlap', 0.23)
set_dict_tuple(DLModelDetectionParam, 'max_overlap_class_agnostic', 0.27)
set_dict_tuple(DLModelDetectionParam, 'min_level', 2)
set_dict_tuple(DLModelDetectionParam, 'max_level', 4)
set_dict_tuple(DLModelDetectionParam, 'anchor_num_subscales', 3)
set_dict_tuple(DLModelDetectionParam, 'anchor_aspect_ratios', [0.534796,0.98496,1.48999,2.24198])
set_dict_tuple(DLModelDetectionParam, 'capacity', 'medium')

***创建模型
Backbone := 'pretrained_dl_classifier_compact.hdl'
* Create the model.
create_dl_model_detection (Backbone, 10, DLModelDetectionParam, DLModelHandle)
DLModelFileName := 'D:/Source/SourceMe/Halcon深度学习模型/Halcon对象检测模型/Dataset' + '/pretrained_dl_model_detection.hdl'
write_dl_model (DLModelHandle, DLModelFileName)

***创建预处理参数
create_dict(DLPreprocessParam)
set_dict_tuple(DLPreprocessParam, 'model_type', 'detection')
set_dict_tuple(DLPreprocessParam, 'image_width', 512)
set_dict_tuple(DLPreprocessParam, 'image_height',320)
set_dict_tuple(DLPreprocessParam, 'image_num_channels',3)
set_dict_tuple(DLPreprocessParam, 'image_range_min',-127)
set_dict_tuple(DLPreprocessParam, 'image_range_max',128)
set_dict_tuple(DLPreprocessParam, 'normalization_type','none')
set_dict_tuple(DLPreprocessParam, 'domain_handling','full_domain')
set_dict_tuple(DLPreprocessParam, 'augmentation', 'false')
set_dict_tuple(DLPreprocessParam, 'instance_type', 'rectangle1')

***创建samples
DLSamples:=[]
NumberImageKeys:=|ImageKeys|
IndexAnnotKeyStart:=0
IndexAnnotKeyEnd:=|AnnotationKeys|
DatasetImageDir:='D:/Source/SourceMe/Halcon深度学习模型/Halcon对象检测模型/Dataset'
DatasetPreprocessDir:='D:/Source/SourceMe/Halcon深度学习模型/Halcon对象检测模型/Dataset/DatasetPreprocess'

***间隔多小显示下
IntervalNumShow:=60 
CurrentIntervalNum:=0

for Index := 0 to NumberImageKeys-1 by 1
    ***获取图片信息
    get_dict_tuple (ImageList, ImageKeys[Index], ImageDict)
    get_dict_tuple (ImageDict, 'file_name', ImageName)
    get_dict_tuple (ImageDict, 'id', ImageID)
    
    ***读取图片
    ImagePath:=DatasetImageDir+'/'+ImageName
    read_image(Image, ImagePath)
    
    ***获取原来图片大小
    get_image_size(Image, Width, Height)
        
    zoom_image_size (Image, Image, 512, 320, 'constant')
    convert_image_type (Image, Image, 'real')
    
    RescaleRange := (128 -(-127) ) / 255.0
    scale_image (Image, Image, RescaleRange, (-127))
             
    * 计算缩放因子
    FactorResampleWidth := real(512) / Width
    FactorResampleHeight := real(320) / Height
    
    create_dict (DLSample)
    set_dict_tuple (DLSample, 'image_id', ImageID)
    set_dict_tuple (DLSample, 'image_file_name', ImageName)

    AnnotBboxRow1s:=[]
    AnnotBboxColumn1s:=[]
    AnnotBboxRow2s:=[]
    AnnotBboxColumn2s:=[]
    AnnotClassIDs:=[] 
    AnnotIndexClass:=0
    Find:=1
    while(Find==1)
        if(IndexAnnotKeyStart<IndexAnnotKeyEnd)
            get_dict_tuple(AnnotationList, AnnotationKeys[IndexAnnotKeyStart], Annotation)
            get_dict_tuple(Annotation, 'image_id', image_id)       
            if(image_id==ImageID)        
                get_dict_tuple(Annotation, 'category_id', category_id)        
                AnnotClassIDs[AnnotIndexClass]:=category_id         
                get_dict_tuple(Annotation, 'bbox', bbox)
                get_dict_tuple(bbox, 0, column) 
                get_dict_tuple(bbox, 1, row)
                get_dict_tuple(bbox, 2, width)
                get_dict_tuple(bbox, 3, height)
                AnnotBboxColumn1s[AnnotIndexClass]:=column*FactorResampleWidth-0.5
                AnnotBboxRow1s[AnnotIndexClass]:=row*FactorResampleHeight-0.5
                AnnotBboxColumn2s[AnnotIndexClass]:=(column+width)*FactorResampleWidth-0.5
                AnnotBboxRow2s[AnnotIndexClass]:=(row+height)*FactorResampleHeight-0.5
                AnnotIndexClass:=AnnotIndexClass+1       
            else
                Find:=0
                ***当没有放入
                 IndexAnnotKeyStart:=IndexAnnotKeyStart-1  
                
            endif
            IndexAnnotKeyStart:=IndexAnnotKeyStart+1   
        else
            Find:=0
        endif
    endwhile
    
    if(CurrentIntervalNum==IntervalNumShow)
        ***显示下预处理的图片    
        dev_display(Image)  
        gen_rectangle1(Rectangle, AnnotBboxRow1s, AnnotBboxColumn1s, AnnotBboxRow2s, AnnotBboxColumn2s)   
        dev_display(Rectangle)
        wait_seconds(1)
        
        CurrentIntervalNum:=-1
    endif
    CurrentIntervalNum:=CurrentIntervalNum+1
    *wait_seconds(1)

    ***设置box跟id
    set_dict_tuple (DLSample, 'bbox_row1', AnnotBboxRow1s)       
    set_dict_tuple (DLSample, 'bbox_col1', AnnotBboxColumn1s)      
    set_dict_tuple (DLSample, 'bbox_row2', AnnotBboxRow2s)      
    set_dict_tuple (DLSample, 'bbox_col2', AnnotBboxColumn2s)       
    set_dict_tuple (DLSample, 'bbox_label_id', AnnotClassIDs)
    set_dict_object(Image, DLSample, 'image')
    * Generate the output file name.
    FileNameOut := ImageID + '_dlsample.hdict'
    RaiseErrorWriteDict := 'true'
    write_dict (DLSample, DatasetPreprocessDir + '/' + FileNameOut, 'raise_error_if_content_not_serializable', RaiseErrorWriteDict)
    * Add output path to DLDataset sample dictionary.
    set_dict_tuple (DLSample, 'dlsample_file_name', FileNameOut)
    * 
    DLSamples[Index]:=DLSample  
    ***显示预处理进度
    dev_disp_text('AllImageKeys:'+NumberImageKeys+'   ImageKey:'+(Index+1), 'window', 10, 10, 'black', 'box', 'true')
endfor

stop()

***开始训练

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

***读取初始模型
*read_dl_model (DLModelFileName, DLModelHandle)


BatchSize:=2
*设置批量
set_dl_model_param (DLModelHandle, 'batch_size', BatchSize)

*设置训练的设备
set_dl_model_param (DLModelHandle, 'device', DLDevice)

***初始化模型学习率
set_dl_model_param(DLModelHandle, 'learning_rate', 0.0001)

***模型训练周期次数
NumEpochs:=100

***数据批次的计数
DatasetSampleIndex:=0

***训练结果的集合
TrainResultsIndex:=0
TrainResults:=[]

***创建一个显示训练的窗口
dev_open_window(0, 0, 512, 512, 'black', WindowHandleTrain)
***创建一个显示信息的窗口
dev_open_window(0, 0, 512, 512, 'black', WindowHandleMessage)

SampleIndicesTrain := []
    
n:= ceil(NumEpochs) 

for Index := 0 to ceil(NumEpochs) - 1 by 1  
    *tuple_shuffle (DatasetSamples, Shuffled)
     
    n:=rand(|DLSamples|)
    n:= sort_index(n)
    
    ShuffleIndices := sort_index(rand(|DLSamples|))
    Shuffled := DLSamples[ShuffleIndices]

    SampleIndicesTrain := [SampleIndicesTrain,Shuffled]
endfor

NumIterationsPerEpoch := int(floor(|DLSamples| / real(BatchSize)))     
NumIterations := int(NumIterationsPerEpoch * NumEpochs)

ComputeMeanLoss:=60

for NumEpoch:=0 to NumIterations by 1

    ***计算时间
    count_seconds(Startime)
    
    ***显示训练信息
    dev_set_window(WindowHandleMessage)
    dev_disp_text('AllNumEpochs:'+NumIterations+' NumEpoch:'+NumEpoch, 'window', 10, 10, 'red', [], [])

    ***判断要不要更换学习率
    
    **批量是BatchSize，读取多小个样品
    DLSampleBatch:=[]
    for DLSamplesProIndex:=0 to BatchSize-1 by 1
       if ((DatasetSampleIndex+DLSamplesProIndex)>=|SampleIndicesTrain|)
           DatasetSampleIndex:=0
       else
           DatasetSampleIndex:=DatasetSampleIndex+DLSamplesProIndex
       endif
       DLSamplesProc := SampleIndicesTrain[DatasetSampleIndex]
       get_dict_tuple (DLSamplesProc, 'dlsample_file_name', DictPath)
       read_dict (DatasetPreprocessDir + '/' + DictPath, [], [], DLSample)
       DLSampleBatch[DLSamplesProIndex]:=DLSample
       
       ***显示训练图片
       dev_set_window(WindowHandleTrain)
       get_dict_object(train_pre_image, DLSampleBatch[DLSamplesProIndex], 'image')
      
       dev_display(train_pre_image)
    endfor
    
    ***当前批次训练模型
    train_dl_model_batch (DLModelHandle, DLSampleBatch, TrainResult)
    ***记录训练的结果
    TrainResults[TrainResultsIndex]:=TrainResult
    
    get_dict_tuple (TrainResult, 'total_loss', LossValue)
    
    ***显示平均loss
    dev_set_window(WindowHandleMessage)    
    ***显示当前loss
    dev_disp_text('total_loss:'+LossValue, 'window', 100, 10, 'red', [], [])
    
    if(TrainResultsIndex==ComputeMeanLoss)
         ***计算平均loss
         LossValues:=[]
         for LossValuesIndex:=0 to TrainResultsIndex by 1
             get_dict_tuple (TrainResults[LossValuesIndex], 'total_loss', LossValue)
             LossValues[LossValuesIndex]:=LossValue
         endfor
         LossMean := mean(LossValues)
         ***显示平均loss
         dev_set_window(WindowHandleMessage)
         ***显示当前loss
         dev_disp_text('mean_loss:'+LossMean, 'window', 200, 10, 'red', [], [])
         TrainResultsIndex:=0
    else
        ***下一次结果加一
        TrainResultsIndex:=TrainResultsIndex+1
    endif

    count_seconds(Endtime)
    t:=Endtime-Startime
    
    all_t:=t*NumIterations    
    use_t:=t*NumEpoch
            
    ***显示时间
    dev_set_window(WindowHandleMessage)        
    ***显示时间
    dev_disp_text('AllTime:'+all_t+'SpeedTime'+use_t, 'window', 300, 10, 'red', [], [])

endfor

stop()
dev_close_window()
dev_close_window()

*优化内存消耗
set_dl_model_param (DLModelHandle, 'optimize_for_inference', 'true')
**优化模型的路径
path_optimize_model_:=DataDirectory+'/optimize_model.hdl'
*重新模型
write_dl_model (DLModelHandle, path_optimize_model_)