|
下面是一段rknn 模型每一层的精度信息
请问 eculidean_norm和eculidean 分别代表原始模型和优化后模型的欧拉距离么,还是有别的含义,我想知道如何对比
我在原始darknet 上看不到有打印每一层精度的函数, 我们的精度信息又没有介绍(文档没有具体的说明,仅仅说明了可以通过accuracy_analysis 分析精度)
======================= Entire Model Quantization Error Analysis =======================
------------------------------------------------------------------------------------------
input_0_out0_nhwc_1_608_608_3.tensor eculidean_norm=0.002969 cosine_norm=0.999995 eculidean=1.168606 cosine=0.999996
convolution_0_1_out0_nhwc_1_304_304_32.tensor eculidean_norm=0.036099 cosine_norm=0.999349 eculidean=180.426132 cosine=0.999348
leakyrelu_0_3_out0_nhwc_1_304_304_32.tensor eculidean_norm=0.065979 cosine_norm=0.997823 eculidean=161.195450 cosine=0.997822
convolution_1_4_out0_nhwc_1_152_152_64.tensor eculidean_norm=0.089029 cosine_norm=0.996037 eculidean=237.020950 cosine=0.996037
leakyrelu_1_6_out0_nhwc_1_152_152_64.tensor eculidean_norm=0.100213 cosine_norm=0.994979 eculidean=194.508972 cosine=0.994979
convolution_2_7_out0_nhwc_1_152_152_64.tensor eculidean_norm=0.093981 cosine_norm=0.995584 eculidean=238.018951 cosine=0.995583
leakyrelu_2_9_out0_nhwc_1_152_152_64.tensor eculidean_norm=0.112215 cosine_norm=0.993704 eculidean=147.640869 cosine=0.993704
slice_route_3_10_out0_nhwc_1_152_152_32.tensor eculidean_norm=0.111622 cosine_norm=0.993770 eculidean=92.506744 cosine=0.993770
convolution_4_11_out0_nhwc_1_152_152_32.tensor eculidean_norm=0.105550 cosine_norm=0.994429 eculidean=186.904648 cosine=0.994430
leakyrelu_4_13_out0_nhwc_1_152_152_32.tensor eculidean_norm=0.120079 cosine_norm=0.992791 eculidean=146.271484 cosine=0.992790
convolution_5_14_out0_nhwc_1_152_152_32.tensor eculidean_norm=0.107417 cosine_norm=0.994231 eculidean=229.388367 cosine=0.994231
leakyrelu_5_16_out0_nhwc_1_152_152_32.tensor eculidean_norm=0.108117 cosine_norm=0.994156 eculidean=192.018097 cosine=0.994156
concat_6_17_out0_nhwc_1_152_152_64.tensor eculidean_norm=0.112150 cosine_norm=0.993711 eculidean=241.384125 cosine=0.993711
convolution_7_18_out0_nhwc_1_152_152_64.tensor eculidean_norm=0.113635 cosine_norm=0.993543 eculidean=478.379272 cosine=0.993544
leakyrelu_7_20_out0_nhwc_1_152_152_64.tensor eculidean_norm=0.190100 cosine_norm=0.981931 eculidean=209.532364 cosine=0.981931
concat_8_21_out0_nhwc_1_152_152_128.tensor eculidean_norm=0.149364 cosine_norm=0.988846 eculidean=256.323303 cosine=0.988845
pooling_9_22_out0_nhwc_1_76_76_128.tensor eculidean_norm=0.140580 cosine_norm=0.990119 eculidean=162.317963 cosine=0.990119
convolution_10_23_out0_nhwc_1_76_76_128.tensor eculidean_norm=0.099966 cosine_norm=0.995003 eculidean=206.569687 cosine=0.995003
leakyrelu_10_25_out0_nhwc_1_76_76_128.tensor eculidean_norm=0.150415 cosine_norm=0.988688 eculidean=92.588127 cosine=0.988688
slice_route_11_26_out0_nhwc_1_76_76_64.tensor eculidean_norm=0.138438 cosine_norm=0.990418 eculidean=61.087040 cosine=0.990417
convolution_12_27_out0_nhwc_1_76_76_64.tensor eculidean_norm=0.114506 cosine_norm=0.993444 eculidean=152.948669 cosine=0.993444
leakyrelu_12_29_out0_nhwc_1_76_76_64.tensor eculidean_norm=0.117270 cosine_norm=0.993124 eculidean=88.808929 cosine=0.993124
convolution_13_30_out0_nhwc_1_76_76_64.tensor eculidean_norm=0.117010 cosine_norm=0.993154 eculidean=176.785812 cosine=0.993154
leakyrelu_13_32_out0_nhwc_1_76_76_64.tensor eculidean_norm=0.125519 cosine_norm=0.992122 eculidean=114.247704 cosine=0.992123
concat_14_33_out0_nhwc_1_76_76_128.tensor eculidean_norm=0.122212 cosine_norm=0.992532 eculidean=144.705093 cosine=0.992532
convolution_15_34_out0_nhwc_1_76_76_128.tensor eculidean_norm=0.099291 cosine_norm=0.995071 eculidean=272.488281 cosine=0.995071
|
|