官方示例:Yolov5s部署
1 验证模型
利用hb_mapper checker后面跟一堆参数来对模型进行配置
2 准备校准数据
因为BPU是INT8计算,所以注定会有精度损失。而且这些误差也是可以传递的,所以到后面精度是越来越低的。如果网络深度过高,也会导致整体精度的下降。
python3 …/…/…/data_preprocess.py
–src_dir …/…/…/01_common/calibration_data/coco
–dst_dir ./calibration_data_rgb_f32
–pic_ext .rgb
–read_mode opencv
3
hb_mapper makertbin --config ./yolov5s_config.yaml --model-type onnx
# 模型转化相关的参数
# ------------------------------------
# model conversion related parameters
model_parameters:
# Onnx浮点网络数据模型文件
# -----------------------------------------------------------
# the model file of floating-point ONNX neural network data
onnx_model: '../../../01_common/model_zoo/mapper/detection/yolov5_onnx_optimized/YOLOv5s.onnx'
# 适用BPU架构
# --------------------------------
# the applicable BPU architecture
march: "bernoulli2"
# 指定模型转换过程中是否输出各层的中间结果,如果为True,则输出所有层的中间输出结果,
# --------------------------------------------------------------------------------------
# specifies whether or not to dump the intermediate results of all layers in conversion
# if set to True, then the intermediate results of all layers shall be dumped
layer_out_dump: False
# 模型转换输出的结果的存放目录
# -----------------------------------------------------------
# the directory in which model conversion results are stored
working_dir: 'model_output'
# 模型转换输出的用于上板执行的模型文件的名称前缀
# -----------------------------------------------------------------------------------------
# model conversion generated name prefix of those model files used for dev board execution
output_model_file_prefix: 'yolov5s_672x672_nv12'
# 模型输入相关参数, 若输入多个节点, 则应使用';'进行分隔, 使用默认缺省设置则写None
# --------------------------------------------------------------------------
# model input related parameters,
# please use ";" to seperate when inputting multiple nodes,
# please use None for default setting
input_parameters:
# (选填) 模型输入的节点名称, 此名称应与模型文件中的名称一致, 否则会报错, 不填则会使用模型文件中的节点名称
# --------------------------------------------------------------------------------------------------------
# (Optional) node name of model input,
# it shall be the same as the name of model file, otherwise an error will be reported,
# the node name of model file will be used when left blank
input_name: ""
# 网络实际执行时,输入给网络的数据格式,包括 nv12/rgb/bgr/yuv444/gray/featuremap,
# ------------------------------------------------------------------------------------------
# the data formats to be passed into neural network when actually performing neural network
# available options: nv12/rgb/bgr/yuv444/gray/featuremap,
input_type_rt: 'nv12'
# 网络实际执行时输入的数据排布, 可选值为 NHWC/NCHW
# 若input_type_rt配置为nv12,则此处参数不需要配置
# ------------------------------------------------------------------
# the data layout formats to be passed into neural network when actually performing neural network, available options: NHWC/NCHW
# If input_type_rt is configured as nv12, then this parameter does not need to be configured
#input_layout_rt: ''
# 网络训练时输入的数据格式,可选的值为rgb/bgr/gray/featuremap/yuv444
# --------------------------------------------------------------------
# the data formats in network training
# available options: rgb/bgr/gray/featuremap/yuv444
input_type_train: 'rgb'
# 网络训练时输入的数据排布, 可选值为 NHWC/NCHW
# ------------------------------------------------------------------
# the data layout in network training, available options: NHWC/NCHW
input_layout_train: 'NCHW'
# (选填) 模型网络的输入大小, 以'x'分隔, 不填则会使用模型文件中的网络输入大小,否则会覆盖模型文件中输入大小
# -------------------------------------------------------------------------------------------
# (Optional)the input size of model network, seperated by 'x'
# note that the network input size of model file will be used if left blank
# otherwise it will overwrite the input size of model file
input_shape: ''
# 网络实际执行时,输入给网络的batch_size, 默认值为1
# ---------------------------------------------------------------------
# the data batch_size to be passed into neural network when actually performing neural network, default value: 1
#input_batch: 1
# 网络输入的预处理方法,主要有以下几种:
# no_preprocess 不做任何操作
# data_mean 减去通道均值mean_value
# data_scale 对图像像素乘以data_scale系数
# data_mean_and_scale 减去通道均值后再乘以scale系数
# -------------------------------------------------------------------------------------------
# preprocessing methods of network input, available options:
# 'no_preprocess' indicates that no preprocess will be made
# 'data_mean' indicates that to minus the channel mean, i.e. mean_value
# 'data_scale' indicates that image pixels to multiply data_scale ratio
# 'data_mean_and_scale' indicates that to multiply scale ratio after channel mean is minused
norm_type: 'data_scale'
# 图像减去的均值, 如果是通道均值,value之间必须用空格分隔
# --------------------------------------------------------------------------
# the mean value minused by image
# note that values must be seperated by space if channel mean value is used
mean_value: ''
# 图像预处理缩放比例,如果是通道缩放比例,value之间必须用空格分隔
# ---------------------------------------------------------------------------
# scale value of image preprocess
# note that values must be seperated by space if channel scale value is used
scale_value: 0.003921568627451
# 模型量化相关参数
# -----------------------------
# model calibration parameters
calibration_parameters:
# 模型量化的参考图像的存放目录,图片格式支持Jpeg、Bmp等格式,输入的图片
# 应该是使用的典型场景,一般是从测试集中选择20~100张图片,另外输入
# 的图片要覆盖典型场景,不要是偏僻场景,如过曝光、饱和、模糊、纯黑、纯白等图片
# 若有多个输入节点, 则应使用';'进行分隔
# -------------------------------------------------------------------------------------------------
# the directory where reference images of model quantization are stored
# image formats include JPEG, BMP etc.
# should be classic application scenarios, usually 20~100 images are picked out from test datasets
# in addition, note that input images should cover typical scenarios
# and try to avoid those overexposed, oversaturated, vague,
# pure blank or pure white images
# use ';' to seperate when there are multiple input nodes
cal_data_dir: './calibration_data_rgb_f32'
# 校准数据二进制文件的数据存储类型,可选值为:float32, uint8
# calibration data binary file save type, available options: float32, uint8
cal_data_type: 'float32'
# 如果输入的图片文件尺寸和模型训练的尺寸不一致时,并且preprocess_on为true,
# 则将采用默认预处理方法(skimage resize),
# 将输入图片缩放或者裁减到指定尺寸,否则,需要用户提前把图片处理为训练时的尺寸
# ---------------------------------------------------------------------------------
# In case the size of input image file is different from that of in model training
# and that preprocess_on is set to True,
# shall the default preprocess method(skimage resize) be used
# i.e., to resize or crop input image into specified size
# otherwise user must keep image size as that of in training in advance
# preprocess_on: False
# 模型量化的算法类型,支持kl、max、default、load,通常采用default即可满足要求, 若为QAT导出的模型, 则应选择load
# ----------------------------------------------------------------------------------
# types of model quantization algorithms, usually default will meet the need
# available options:kl, max, default and load
# if converted model is quanti model exported from QAT , then choose `load`
calibration_type: 'default'
# 编译器相关参数
# ----------------------------
# compiler related parameters
compiler_parameters:
# 编译策略,支持bandwidth和latency两种优化模式;
# bandwidth以优化ddr的访问带宽为目标;
# latency以优化推理时间为目标
# -------------------------------------------------------------------------------------------
# compilation strategy, there are 2 available optimization modes: 'bandwidth' and 'lantency'
# the 'bandwidth' mode aims to optimize ddr access bandwidth
# while the 'lantency' mode aims to optimize inference duration
compile_mode: 'latency'
# 设置debug为True将打开编译器的debug模式,能够输出性能仿真的相关信息,如帧率、DDR带宽占用等
# -----------------------------------------------------------------------------------
# the compiler's debug mode will be enabled by setting to True
# this will dump performance simulation related information
# such as: frame rate, DDR bandwidth usage etc.
debug: False
# 编译模型指定核数,不指定默认编译单核模型, 若编译双核模型,将下边注释打开即可
# -------------------------------------------------------------------------------------
# specifies number of cores to be used in model compilation
# as default, single core is used as this value left blank
# please delete the "# " below to enable dual-core mode when compiling dual-core model
# core_num: 2
[root@27422ea0f8ea mapper]# hb_mapper makertbin --config ./yolov5s_config.yaml --model-type onnx
2022-10-04 17:45:17,252 INFO Start hb_mapper....
2022-10-04 17:45:17,252 INFO log will be stored in /open_explorer/ddk/samples/ai_toolchain/horizon_model_convert_sample/04_detection/03_yolov5s/mapper/hb_mapper_makertbin.log
2022-10-04 17:45:17,252 INFO hbdk version 3.37.2
2022-10-04 17:45:17,252 INFO horizon_nn version 0.14.0
2022-10-04 17:45:17,252 INFO hb_mapper version 1.9.9
2022-10-04 17:45:17,252 INFO Start Model Convert....
2022-10-04 17:45:17,263 INFO Using abs path /open_explorer/ddk/samples/ai_toolchain/horizon_model_convert_sample/01_common/model_zoo/mapper/detection/yolov5_onnx_optimized/YOLOv5s.onnx
2022-10-04 17:45:17,264 INFO validating model_parameters...
2022-10-04 17:45:17,283 WARNING User input 'log_level' deleted,Please do not use this parameter again
2022-10-04 17:45:17,283 INFO Using abs path /open_explorer/ddk/samples/ai_toolchain/horizon_model_convert_sample/04_detection/03_yolov5s/mapper/model_output
2022-10-04 17:45:17,283 INFO validating model_parameters finished
2022-10-04 17:45:17,284 INFO validating input_parameters...
2022-10-04 17:45:17,284 INFO input num is set to 1 according to input_names
2022-10-04 17:45:17,284 INFO model name missing, using model name from model file: ['data']
2022-10-04 17:45:17,284 INFO model input shape missing, using shape from model file: [[1, 3, 672, 672]]
2022-10-04 17:45:17,284 INFO validating input_parameters finished
2022-10-04 17:45:17,284 INFO validating calibration_parameters...
2022-10-04 17:45:17,284 INFO Using abs path /open_explorer/ddk/samples/ai_toolchain/horizon_model_convert_sample/04_detection/03_yolov5s/mapper/calibration_data_rgb_f32
2022-10-04 17:45:17,284 INFO validating calibration_parameters finished
2022-10-04 17:45:17,284 INFO validating custom_op...
2022-10-04 17:45:17,284 INFO custom_op does not exist, skipped
2022-10-04 17:45:17,284 INFO validating custom_op finished
2022-10-04 17:45:17,285 INFO validating compiler_parameters...
2022-10-04 17:45:17,285 INFO validating compiler_parameters finished
2022-10-04 17:45:17,288 INFO The calibration dir name suffix is the same as the value float32 of the cal_data_type parameter and will be read with the value of cal_data_type
2022-10-04 17:45:17,288 INFO *******************************************
2022-10-04 17:45:17,288 INFO First calibration picture name: COCO_val2014_000000181007.rgb
2022-10-04 17:45:17,288 INFO First calibration picture md5:
136bb23027c812cc2978395421fe6be7 /open_explorer/ddk/samples/ai_toolchain/horizon_model_convert_sample/04_detection/03_yolov5s/mapper/calibration_data_rgb_f32/COCO_val2014_000000181007.rgb
2022-10-04 17:45:17,302 INFO *******************************************
2022-10-04 17:45:20,374 INFO [Tue Oct 4 17:45:20 2022] Start to Horizon NN Model Convert.
2022-10-04 17:45:20,374 INFO Parsing the input parameter:{'data': {'input_shape': [1, 3, 672, 672], 'expected_input_type': 'YUV444_128', 'original_input_type': 'RGB', 'original_input_layout': 'NCHW', 'scales': array([0.00392157], dtype=float32)}}
2022-10-04 17:45:20,375 INFO Parsing the calibration parameter
2022-10-04 17:45:20,375 INFO Parsing the hbdk parameter:{'hbdk_pass_through_params': '--fast --O3', 'input-source': {'data': 'pyramid', '_default_value': 'ddr'}}
2022-10-04 17:45:20,375 INFO HorizonNN version: 0.14.0
2022-10-04 17:45:20,375 INFO HBDK version: 3.37.2
2022-10-04 17:45:20,375 INFO [Tue Oct 4 17:45:20 2022] Start to parse the onnx model.
2022-10-04 17:45:20,383 INFO Input ONNX model infomation:
ONNX IR version: 6
Opset version: 10
Producer: pytorch1.6
Domain: none
Input name: data, [1, 3, 672, 672]
Output name: output, [1, 84, 84, 255]
Output name: 641, [1, 42, 42, 255]
Output name: 643, [1, 21, 21, 255]
2022-10-04 17:45:20,448 INFO [Tue Oct 4 17:45:20 2022] End to parse the onnx model.
2022-10-04 17:45:20,448 INFO Model input names: ['data']
2022-10-04 17:45:20,449 INFO Create a preprocessing operator for input_name data with means=None, std=[254.99998492], original_input_layout=NCHW, color convert from 'RGB' to 'YUV_BT601_FULL_RANGE'.
2022-10-04 17:45:20,506 INFO Saving the original float model: yolov5s_672x672_nv12_original_float_model.onnx.
2022-10-04 17:45:20,507 INFO [Tue Oct 4 17:45:20 2022] Start to optimize the model.
2022-10-04 17:45:21,519 INFO [Tue Oct 4 17:45:21 2022] End to optimize the model.
2022-10-04 17:45:21,537 INFO Saving the optimized model: yolov5s_672x672_nv12_optimized_float_model.onnx.
2022-10-04 17:45:21,537 INFO [Tue Oct 4 17:45:21 2022] Start to calibrate the model.
2022-10-04 17:45:21,537 INFO There are 50 samples in the calibration data set.
2022-10-04 17:45:21,952 INFO Run calibration model with default calibration method.
2022-10-04 17:46:17,313 INFO Select max-percentile:percentile=0.99995 method.
2022-10-04 17:46:17,348 INFO [Tue Oct 4 17:46:17 2022] End to calibrate the model.
2022-10-04 17:46:17,348 INFO [Tue Oct 4 17:46:17 2022] Start to quantize the model.
2022-10-04 17:46:22,590 INFO input data is from pyramid. Its layout is set to NHWC
2022-10-04 17:46:22,796 INFO [Tue Oct 4 17:46:22 2022] End to quantize the model.
2022-10-04 17:46:22,962 INFO Saving the quantized model: yolov5s_672x672_nv12_quantized_model.onnx.
2022-10-04 17:46:23,495 INFO [Tue Oct 4 17:46:23 2022] Start to compile the model with march bernoulli2.
2022-10-04 17:46:23,743 INFO Compile submodel: torch-jit-export_subgraph_0
2022-10-04 17:46:24,120 INFO hbdk-cc parameters:['--fast', '--O3', '--input-layout', 'NHWC', '--output-layout', 'NHWC', '--input-source', 'pyramid']
2022-10-04 17:46:24,167 INFO INFO: "-j" or "--jobs" is not specified, launch 16 threads for optimization
2022-10-04 17:46:24,167 WARNING missing stride for pyramid input[0], use its aligned width by default.
[==================================================] 100%
2022-10-04 17:47:42,398 INFO consumed time 78.234
2022-10-04 17:47:42,572 INFO FPS=17.43, latency = 57387.6 us (see torch-jit-export_subgraph_0.html)
2022-10-04 17:47:42,746 INFO [Tue Oct 4 17:47:42 2022] End to compile the model with march bernoulli2.
2022-10-04 17:47:42,748 INFO The converted model node information:
============================================================================================================================================
Node ON Subgraph Type Cosine Similarity Threshold
--------------------------------------------------------------------------------------------------------------------------------------------
HZ_PREPROCESS_FOR_data BPU id(0) HzSQuantizedPreprocess 1.000249 127.000000
Slice_4 BPU id(0) Slice 0.999995 1.065925
Slice_9 BPU id(0) Slice 0.999896 1.065925
Slice_14 BPU id(0) Slice 0.999992 1.065925
Slice_19 BPU id(0) Slice 0.999895 1.065925
Slice_24 BPU id(0) Slice 0.999995 1.065925
Slice_29 BPU id(0) Slice 0.999894 1.065925
Slice_34 BPU id(0) Slice 0.999992 1.065925
Slice_39 BPU id(0) Slice 0.999894 1.065925
Concat_40 BPU id(0) Concat 1.000201 1.065925
Conv_41 BPU id(0) HzSQuantizedConv 1.000417 1.065925
LeakyRelu_43 BPU id(0) HzLeakyRelu 0.999315 9.565388
Conv_44 BPU id(0) HzSQuantizedConv 0.998291 9.565388
LeakyRelu_46 BPU id(0) HzLeakyRelu 0.998607 13.493938
Conv_47 BPU id(0) HzSQuantizedConv 0.999357 13.493938
LeakyRelu_49 BPU id(0) HzLeakyRelu 0.999475 4.795215
Conv_50 BPU id(0) HzSQuantizedConv 0.993676 4.795215
LeakyRelu_52 BPU id(0) HzLeakyRelu 0.997127 7.377336
Conv_53 BPU id(0) HzSQuantizedConv 0.995438 7.377336
LeakyRelu_55 BPU id(0) HzLeakyRelu 0.996490 12.893383
UNIT_CONV_FOR_Add_56 BPU id(0) HzSQuantizedConv 0.997635 4.795215
Conv_57 BPU id(0) HzSQuantizedConv 0.993456 13.701491
Conv_58 BPU id(0) HzSQuantizedConv 0.995226 13.493938
Concat_59 BPU id(0) Concat 0.994014 11.856925
LeakyRelu_61 BPU id(0) HzLeakyRelu 0.995508 11.856925
Conv_62 BPU id(0) HzSQuantizedConv 0.991005 11.856925
LeakyRelu_64 BPU id(0) HzLeakyRelu 0.993757 7.449783
Conv_65 BPU id(0) HzSQuantizedConv 0.990448 7.449783
LeakyRelu_67 BPU id(0) HzLeakyRelu 0.992342 7.031975
Conv_68 BPU id(0) HzSQuantizedConv 0.997501 7.031975
LeakyRelu_70 BPU id(0) HzLeakyRelu 0.998094 3.502887
Conv_71 BPU id(0) HzSQuantizedConv 0.994286 3.502887
LeakyRelu_73 BPU id(0) HzLeakyRelu 0.994844 6.069211
Conv_74 BPU id(0) HzSQuantizedConv 0.992879 6.069211
LeakyRelu_76 BPU id(0) HzLeakyRelu 0.993573 4.123940
UNIT_CONV_FOR_Add_77 BPU id(0) HzSQuantizedConv 0.997019 3.502887
Conv_78 BPU id(0) HzSQuantizedConv 0.989430 5.291677
LeakyRelu_80 BPU id(0) HzLeakyRelu 0.985459 5.891102
Conv_81 BPU id(0) HzSQuantizedConv 0.989856 5.891102
LeakyRelu_83 BPU id(0) HzLeakyRelu 0.990840 7.603921
UNIT_CONV_FOR_Add_84 BPU id(0) HzSQuantizedConv 0.995930 5.291677
Conv_85 BPU id(0) HzSQuantizedConv 0.992432 8.904430
LeakyRelu_87 BPU id(0) HzLeakyRelu 0.993280 5.469590
Conv_88 BPU id(0) HzSQuantizedConv 0.989034 5.469590
LeakyRelu_90 BPU id(0) HzLeakyRelu 0.990263 10.865228
UNIT_CONV_FOR_Add_91 BPU id(0) HzSQuantizedConv 0.995579 8.904430
Conv_92 BPU id(0) HzSQuantizedConv 0.988808 13.730002
Conv_93 BPU id(0) HzSQuantizedConv 0.990489 7.031975
Concat_94 BPU id(0) Concat 0.989560 7.262249
LeakyRelu_96 BPU id(0) HzLeakyRelu 0.990888 7.262249
Conv_97 BPU id(0) HzSQuantizedConv 0.993098 7.262249
LeakyRelu_99 BPU id(0) HzLeakyRelu 0.991211 5.496812
Conv_100 BPU id(0) HzSQuantizedConv 0.993583 5.496812
LeakyRelu_102 BPU id(0) HzLeakyRelu 0.991866 5.719972
Conv_103 BPU id(0) HzSQuantizedConv 0.995982 5.719972
LeakyRelu_105 BPU id(0) HzLeakyRelu 0.996843 2.621932
Conv_106 BPU id(0) HzSQuantizedConv 0.994463 2.621932
LeakyRelu_108 BPU id(0) HzLeakyRelu 0.994016 5.975686
Conv_109 BPU id(0) HzSQuantizedConv 0.995599 5.975686
LeakyRelu_111 BPU id(0) HzLeakyRelu 0.993798 3.731829
UNIT_CONV_FOR_Add_112 BPU id(0) HzSQuantizedConv 0.995422 2.621932
Conv_113 BPU id(0) HzSQuantizedConv 0.994725 4.103075
LeakyRelu_115 BPU id(0) HzLeakyRelu 0.990147 5.049542
Conv_116 BPU id(0) HzSQuantizedConv 0.993896 5.049542
LeakyRelu_118 BPU id(0) HzLeakyRelu 0.992818 6.516913
UNIT_CONV_FOR_Add_119 BPU id(0) HzSQuantizedConv 0.994551 4.103075
Conv_120 BPU id(0) HzSQuantizedConv 0.994880 7.711994
LeakyRelu_122 BPU id(0) HzLeakyRelu 0.990902 4.603847
Conv_123 BPU id(0) HzSQuantizedConv 0.992633 4.603847
LeakyRelu_125 BPU id(0) HzLeakyRelu 0.992069 10.031747
UNIT_CONV_FOR_Add_126 BPU id(0) HzSQuantizedConv 0.994463 7.711994
Conv_127 BPU id(0) HzSQuantizedConv 0.992736 11.763662
Conv_128 BPU id(0) HzSQuantizedConv 0.992591 5.719972
Concat_129 BPU id(0) Concat 0.992644 5.999934
LeakyRelu_131 BPU id(0) HzLeakyRelu 0.992431 5.999934
Conv_132 BPU id(0) HzSQuantizedConv 0.994696 5.999934
LeakyRelu_134 BPU id(0) HzLeakyRelu 0.991275 5.092212
Conv_135 BPU id(0) HzSQuantizedConv 0.995734 5.092212
LeakyRelu_137 BPU id(0) HzLeakyRelu 0.993229 4.080043
Conv_138 BPU id(0) HzSQuantizedConv 0.996398 4.080043
LeakyRelu_140 BPU id(0) HzLeakyRelu 0.997200 7.308180
MaxPool_141 BPU id(0) HzQuantizedMaxPool 0.998847 7.308180
MaxPool_142 BPU id(0) HzQuantizedMaxPool 0.999095 7.308180
MaxPool_143 BPU id(0) HzQuantizedMaxPool 0.999225 7.308180
Concat_144 BPU id(0) Concat 0.998973 7.308180
Conv_145 BPU id(0) HzSQuantizedConv 0.997401 7.308180
LeakyRelu_147 BPU id(0) HzLeakyRelu 0.991993 3.540069
Conv_148 BPU id(0) HzSQuantizedConv 0.995313 3.540069
LeakyRelu_150 BPU id(0) HzLeakyRelu 0.991999 4.885417
Conv_151 BPU id(0) HzSQuantizedConv 0.993966 4.885417
LeakyRelu_153 BPU id(0) HzLeakyRelu 0.991915 4.912245
Conv_154 BPU id(0) HzSQuantizedConv 0.995375 4.912245
LeakyRelu_156 BPU id(0) HzLeakyRelu 0.991952 4.840213
Conv_157 BPU id(0) HzSQuantizedConv 0.993622 4.840213
Conv_158 BPU id(0) HzSQuantizedConv 0.993671 3.540069
Concat_159 BPU id(0) Concat 0.993632 4.888761
LeakyRelu_161 BPU id(0) HzLeakyRelu 0.988410 4.888761
Conv_162 BPU id(0) HzSQuantizedConv 0.992833 4.888761
LeakyRelu_164 BPU id(0) HzLeakyRelu 0.986227 4.201017
Conv_165 BPU id(0) HzSQuantizedConv 0.990234 4.201017
LeakyRelu_167 BPU id(0) HzLeakyRelu 0.988071 5.000985
Resize_168 BPU id(0) HzQuantizedResizeUpsample 0.988088 5.000985
UNIT_CONV_FOR_505_0.039377834647894_TO_FUSE_SCALE BPU id(0) HzSQuantizedConv
Concat_169 BPU id(0) Concat 0.989407 5.000985
Conv_170 BPU id(0) HzSQuantizedConv 0.993669 5.000985
LeakyRelu_172 BPU id(0) HzLeakyRelu 0.993927 3.854206
Conv_173 BPU id(0) HzSQuantizedConv 0.993419 3.854206
LeakyRelu_175 BPU id(0) HzLeakyRelu 0.991511 4.369511
Conv_176 BPU id(0) HzSQuantizedConv 0.991438 4.369511
LeakyRelu_178 BPU id(0) HzLeakyRelu 0.991134 5.215519
Conv_179 BPU id(0) HzSQuantizedConv 0.989505 5.215519
Conv_180 BPU id(0) HzSQuantizedConv 0.989471 5.000985
Concat_181 BPU id(0) Concat 0.989477 5.397383
LeakyRelu_183 BPU id(0) HzLeakyRelu 0.988263 5.397383
Conv_184 BPU id(0) HzSQuantizedConv 0.989288 5.397383
LeakyRelu_186 BPU id(0) HzLeakyRelu 0.985086 5.141273
Conv_187 BPU id(0) HzSQuantizedConv 0.989487 5.141273
LeakyRelu_189 BPU id(0) HzLeakyRelu 0.992006 6.019610
Resize_190 BPU id(0) HzQuantizedResizeUpsample 0.992025 6.019610
UNIT_CONV_FOR_470_0.047398507595062_TO_FUSE_SCALE BPU id(0) HzSQuantizedConv
Concat_191 BPU id(0) Concat 0.991818 6.019610
Conv_192 BPU id(0) HzSQuantizedConv 0.993551 6.019610
LeakyRelu_194 BPU id(0) HzLeakyRelu 0.996678 3.829432
Conv_195 BPU id(0) HzSQuantizedConv 0.994650 3.829432
LeakyRelu_197 BPU id(0) HzLeakyRelu 0.996992 3.844464
Conv_198 BPU id(0) HzSQuantizedConv 0.994258 3.844464
LeakyRelu_200 BPU id(0) HzLeakyRelu 0.996183 5.467257
Conv_201 BPU id(0) HzSQuantizedConv 0.991097 5.467257
Conv_202 BPU id(0) HzSQuantizedConv 0.991353 6.019610
Concat_203 BPU id(0) Concat 0.991187 5.914804
LeakyRelu_205 BPU id(0) HzLeakyRelu 0.993002 5.914804
Conv_206 BPU id(0) HzSQuantizedConv 0.980396 5.914804
LeakyRelu_208 BPU id(0) HzLeakyRelu 0.983803 19.009319
Conv_209 BPU id(0) HzSQuantizedConv 0.981550 19.009319
LeakyRelu_211 BPU id(0) HzLeakyRelu 0.983346 6.489002
UNIT_CONV_FOR_568_0.051094502210617_TO_FUSE_SCALE BPU id(0) HzSQuantizedConv
Concat_212 BPU id(0) Concat 0.988376 6.489002
Conv_213 BPU id(0) HzSQuantizedConv 0.982762 6.489002
LeakyRelu_215 BPU id(0) HzLeakyRelu 0.982941 6.002916
Conv_216 BPU id(0) HzSQuantizedConv 0.983863 6.002916
LeakyRelu_218 BPU id(0) HzLeakyRelu 0.980998 5.994791
Conv_219 BPU id(0) HzSQuantizedConv 0.985687 5.994791
LeakyRelu_221 BPU id(0) HzLeakyRelu 0.985396 6.025105
Conv_222 BPU id(0) HzSQuantizedConv 0.981646 6.025105
Conv_223 BPU id(0) HzSQuantizedConv 0.985091 6.489002
Concat_224 BPU id(0) Concat 0.983152 7.844817
LeakyRelu_226 BPU id(0) HzLeakyRelu 0.985835 7.844817
Conv_227 BPU id(0) HzSQuantizedConv 0.976176 7.844817
LeakyRelu_229 BPU id(0) HzLeakyRelu 0.977843 18.134787
Conv_230 BPU id(0) HzSQuantizedConv 0.980275 18.134787
LeakyRelu_232 BPU id(0) HzLeakyRelu 0.980287 6.185248
UNIT_CONV_FOR_538_0.048702739179134_TO_FUSE_SCALE BPU id(0) HzSQuantizedConv
Concat_233 BPU id(0) Concat 0.982942 6.185248
Conv_234 BPU id(0) HzSQuantizedConv 0.982924 6.185248
LeakyRelu_236 BPU id(0) HzLeakyRelu 0.982356 6.477486
Conv_237 BPU id(0) HzSQuantizedConv 0.979146 6.477486
LeakyRelu_239 BPU id(0) HzLeakyRelu 0.977248 8.212425
Conv_240 BPU id(0) HzSQuantizedConv 0.982827 8.212425
LeakyRelu_242 BPU id(0) HzLeakyRelu 0.980781 6.408384
Conv_243 BPU id(0) HzSQuantizedConv 0.977600 6.408384
Conv_244 BPU id(0) HzSQuantizedConv 0.984553 6.185248
Concat_245 BPU id(0) Concat 0.980870 6.350356
LeakyRelu_247 BPU id(0) HzLeakyRelu 0.983271 6.350356
Conv_248 BPU id(0) HzSQuantizedConv 0.975667 6.350356
LeakyRelu_250 BPU id(0) HzLeakyRelu 0.978477 13.301773
Conv_251 BPU id(0) HzSQuantizedConv 0.999065 19.009319
Conv_253 BPU id(0) HzSQuantizedConv 0.999060 18.134787
Conv_255 BPU id(0) HzSQuantizedConv 0.999241 13.301773
2022-10-04 17:47:42,749 INFO [Tue Oct 4 17:47:42 2022] End to Horizon NN Model Convert.
2022-10-04 17:47:42,829 INFO start convert to *.bin file....
2022-10-04 17:47:42,873 INFO ONNX model output num : 3
2022-10-04 17:47:42,873 INFO ############# model deps info #############
2022-10-04 17:47:42,874 INFO hb_mapper version : 1.9.9
2022-10-04 17:47:42,874 INFO hbdk version : 3.37.2
2022-10-04 17:47:42,874 INFO hbdk runtime version: 3.14.14
2022-10-04 17:47:42,874 INFO horizon_nn version : 0.14.0
2022-10-04 17:47:42,874 INFO ############# model_parameters info #############
2022-10-04 17:47:42,874 INFO onnx_model : /open_explorer/ddk/samples/ai_toolchain/horizon_model_convert_sample/01_common/model_zoo/mapper/detection/yolov5_onnx_optimized/YOLOv5s.onnx
2022-10-04 17:47:42,874 INFO BPU march : bernoulli2
2022-10-04 17:47:42,874 INFO layer_out_dump : False
2022-10-04 17:47:42,874 INFO log_level : DEBUG
2022-10-04 17:47:42,874 INFO working dir : /open_explorer/ddk/samples/ai_toolchain/horizon_model_convert_sample/04_detection/03_yolov5s/mapper/model_output
2022-10-04 17:47:42,874 INFO output_model_file_prefix: yolov5s_672x672_nv12
2022-10-04 17:47:42,874 INFO ############# input_parameters info #############
2022-10-04 17:47:42,874 INFO ------------------------------------------
2022-10-04 17:47:42,874 INFO ---------input info : data ---------
2022-10-04 17:47:42,874 INFO input_name : data
2022-10-04 17:47:42,874 INFO input_type_rt : nv12
2022-10-04 17:47:42,875 INFO input_space&range : regular
2022-10-04 17:47:42,875 INFO input_layout_rt : None
2022-10-04 17:47:42,875 INFO input_type_train : rgb
2022-10-04 17:47:42,875 INFO input_layout_train : NCHW
2022-10-04 17:47:42,875 INFO norm_type : data_scale
2022-10-04 17:47:42,875 INFO input_shape : 1x3x672x672
2022-10-04 17:47:42,875 INFO scale_value : 0.003921568627451,
2022-10-04 17:47:42,875 INFO cal_data_dir : /open_explorer/ddk/samples/ai_toolchain/horizon_model_convert_sample/04_detection/03_yolov5s/mapper/calibration_data_rgb_f32
2022-10-04 17:47:42,875 INFO ---------input info : data end -------
2022-10-04 17:47:42,875 INFO ------------------------------------------
2022-10-04 17:47:42,875 INFO ############# calibration_parameters info #############
2022-10-04 17:47:42,875 INFO preprocess_on : False
2022-10-04 17:47:42,875 INFO calibration_type: : default
2022-10-04 17:47:42,875 INFO cal_data_type : float32
2022-10-04 17:47:42,875 INFO ############# compiler_parameters info #############
2022-10-04 17:47:42,875 INFO hbdk_pass_through_params: --fast --O3
2022-10-04 17:47:42,875 INFO input-source : {'data': 'pyramid', '_default_value': 'ddr'}
2022-10-04 17:47:42,884 INFO Convert to runtime bin file sucessfully!
2022-10-04 17:47:42,884 INFO End Model Convert
REF
https://developer.horizon.ai/forumDetail/107952931390742029