专栏算法工具链板端结果

板端结果

已解决
我爱邢宝宝i2024-04-24
107
4

用户您好，请详细描述您所遇到的问题，详细的描述有助于帮助我们快速定位，解决问题~Thanks♪(･ω･)ﾉ


[model name]: resnet50_3dim


input[0]:

name: input

input source: HB_DNN_INPUT_FROM_PYRAMID

valid shape: (1,3,224,224,)

aligned shape: (1,3,224,224,)

aligned byte size: 75264

tensor type: HB_DNN_IMG_TYPE_NV12

tensor layout: HB_DNN_LAYOUT_NCHW

quanti type: NONE

stride: (0,0,0,0,)


output[0]:

name: output

valid shape: (1,1,1,1000,)

aligned shape: (1,1,1,1024,)

aligned byte size: 1024

tensor type: HB_DNN_TENSOR_TYPE_S8

tensor layout: HB_DNN_LAYOUT_NONE

quanti type: SCALE

stride: (1024,1024,1024,1,)

scale data: 10.1449,


算法工具链
征程5
评论4
0/1000
  • 我爱邢宝宝i
    Lv.2

    您好，请问scale data的含义是什么


    2024-04-24
    0
    0
  • J6标定问题
    Lv.1

    scale data称为量化比例因子或量化系数。在模型反量化中应用到，可以减少模型推理推理时间

    2024-04-24
    0
    0
  • 我爱邢宝宝i
    Lv.2

    ./hrt_model_exec infer --model_file ../test_OE/resnet50_3dim.bin --core_id 1 --input_file ../input/processed_input.jpg --ena

    ble_dump true --hybrid_dequantize_process true --dump_format txt我这样保存的推理结果和PC端差距很大

    2024-04-24
    0
    0
  • J6标定问题
    Lv.1

    ./hrt_model_exec-60 infer --model_file ../mobilenet_test/mobilenetv1

    _224x224_nv12.bin --input_file ../mobilenet_test/zebra_cls.jpg --enable_dump true --hybrid_dequantize_p

    rocess true --dump_format txt --enable_cls_post_process true

    ./hrt_model_exec-60 infer --model_file ../mobilenet_test/mobilenetv1_224x224_nv12.bin --input_file ../mobilenet_test/zebra_cls.jpg

    --enable_dump true --hybrid_dequantize_process true --dump_format txt --enable_cls_post_process true

    I0000 00:00:00.000000 8023 vlog_is_on.cc:197] RAW: Set VLOG level for "*" to 3

    [BPU_PLAT]BPU Platform Version(1.3.3)!

    [HBRT] set log level as 0. version = 3.15.33.0

    [DNN] Runtime version = 1.20.2_(3.15.33 HBRT)

    [A][DNN][packed_model.cpp:246][Model](2000-01-01,08:20:43.266.28) [HorizonRT] The model builder version = 1.11.2

    Load model to DDR cost 367.373ms.

    I0101 08:20:43.314006 8023 function_util.cpp:116] get model handle success

    I0101 08:20:43.314067 8023 function_util.cpp:122] get model input count success

    I0101 08:20:43.314105 8023 function_util.cpp:128] get model output count success

    I0101 08:20:43.314253 8023 function_util.cpp:154] prepare output tensor success

    I0101 08:20:43.339064 8023 function_util.cpp:170] read file success!

    I0101 08:20:43.339299 8023 function_util.cpp:205] infer success

    I0101 08:20:43.341977 8023 function_util.cpp:211] task done


    ---------------------Frame 0 begin---------------------

    Infer time: 2.844 ms

    ---------------------Frame 0 end---------------------

    I0101 08:20:43.342145 8023 function_util.cpp:1289] offset: 0

    I0101 08:20:43.342183 8023 function_util.cpp:1290] element_of_num: 75264

    I0101 08:20:43.342217 8023 function_util.cpp:1291] element_of_memsize: 75264

    I0101 08:20:43.342260 8023 function_util.cpp:1316] dump_txt_axis: -1; aligned_shape.numDimensions: 4; update d ump_txt_axis: 4

    I0101 08:20:43.342296 8023 function_util.cpp:1330] FLAGS_dump_txt_axis: 4

    I0101 08:20:43.342330 8023 function_util.cpp:1332] dump_line_length: 1

    I0101 08:20:44.581587 8023 function_util.cpp:1365] Dump path: ./model_infer_input_0.txt

    I0101 08:20:44.581961 8023 function_util.cpp:968] FLAGS_dump_txt_axis: 4

    I0101 08:20:44.582010 8023 function_util.cpp:970] dump_line_length: 1

    I0101 08:20:44.604005 8023 function_util.cpp:1009] Dump path: ./model_infer_output_0.txt

    I0101 08:20:44.604138 8023 function_util.cpp:248] class result: [id]340, [score]0.986489

    I0101 08:20:44.604187 8023 function_util.cpp:251] Attention: please make sure model is for classfication


    我尝试了mobilenet，结果是ok的

    2024-04-25
    0
    0