打开微信,使用扫一扫进入页面后,点击右上角菜单,
点击“发送给朋友”或“分享到朋友圈”完成分享
我在发现yolov5的逐层和cpu模式结果不太一致后,使用了dump工具
下面是结果打印
Warning: 'replace=True' Op [] _00000_model.0.conv_in errRate = 0.0 _00000_model.0.conv_out errRate = 0.033778149634599686 _00001_model.0.act_in errRate = 0.033778149634599686 _00001_model.0.act_out errRate = 0.03537113219499588 _00002_model.1.conv_in errRate = 0.03537113219499588 ******************* _00002_model.1.conv_out errRate = 0.06306491792201996 ******************* _00003_model.1.act_in errRate = 0.06306491792201996 _00003_model.1.act_out errRate = 0.07197665423154831 _00004_model.2.cv1.conv_in errRate = 0.07197665423154831 _00004_model.2.cv1.conv_out errRate = 0.06885150820016861 _00005_model.2.cv1.act_in errRate = 0.06885150820016861 _00005_model.2.cv1.act_out errRate = 0.0884501039981842 _00006_model.2.m.0.cv1.conv_in errRate = 0.0884501039981842 ******************* _00006_model.2.m.0.cv1.conv_out errRate = 0.15424127876758575 ******************* _00007_model.2.m.0.cv1.act_in errRate = 0.15424127876758575 _00007_model.2.m.0.cv1.act_out errRate = 0.16953909397125244 _00008_model.2.m.0.cv2.conv_in errRate = 0.16953909397125244 _00008_model.2.m.0.cv2.conv_out errRate = 0.137769877910614 _00009_model.2.m.0.cv2.act_in errRate = 0.137769877910614 _00009_model.2.m.0.cv2.act_out errRate = 0.12482896447181702 _00010_model.2.cv2.conv_in errRate = 0.07197665423154831 _00010_model.2.cv2.conv_out errRate = 0.0748973935842514 _00011_model.2.cv2.act_in errRate = 0.0748973935842514 _00011_model.2.cv2.act_out errRate = 0.08865470439195633 _00012_model.2.cv3.conv_in errRate = 0.10123926401138306 _00012_model.2.cv3.conv_out errRate = 0.1390291303396225 _00013_model.2.cv3.act_in errRate = 0.1390291303396225 _00013_model.2.cv3.act_out errRate = 0.15163809061050415 _00014_model.3.conv_in errRate = 0.15163809061050415 _00014_model.3.conv_out errRate = 0.12521281838417053 _00015_model.3.act_in errRate = 0.12521281838417053 _00015_model.3.act_out errRate = 0.11929642409086227 _00016_model.4.cv1.conv_in errRate = 0.11929642409086227 _00016_model.4.cv1.conv_out errRate = 0.09127743542194366 _00017_model.4.cv1.act_in errRate = 0.09127743542194366 _00017_model.4.cv1.act_out errRate = 0.08188644796609879 _00018_model.4.m.0.cv1.conv_in errRate = 0.08188644796609879 _00018_model.4.m.0.cv1.conv_out errRate = 0.12145888805389404 _00019_model.4.m.0.cv1.act_in errRate = 0.12145888805389404 _00019_model.4.m.0.cv1.act_out errRate = 0.11850713938474655 _00020_model.4.m.0.cv2.conv_in errRate = 0.11850713938474655 _00020_model.4.m.0.cv2.conv_out errRate = 0.11148262768983841 _00021_model.4.m.0.cv2.act_in errRate = 0.11148262768983841 _00021_model.4.m.0.cv2.act_out errRate = 0.1129852756857872 _00022_model.4.m.1.cv1.conv_in errRate = 0.11095835268497467 _00022_model.4.m.1.cv1.conv_out errRate = 0.11369039118289948 _00023_model.4.m.1.cv1.act_in errRate = 0.11369039118289948 _00023_model.4.m.1.cv1.act_out errRate = 0.11194079369306564 _00024_model.4.m.1.cv2.conv_in errRate = 0.11194079369306564 _00024_model.4.m.1.cv2.conv_out errRate = 0.11065706610679626 _00025_model.4.m.1.cv2.act_in errRate = 0.11065706610679626 _00025_model.4.m.1.cv2.act_out errRate = 0.10952870547771454 _00026_model.4.cv2.conv_in errRate = 0.11929642409086227 _00026_model.4.cv2.conv_out errRate = 0.15524344146251678 _00027_model.4.cv2.act_in errRate = 0.15524344146251678 _00027_model.4.cv2.act_out errRate = 0.15800440311431885 _00028_model.4.cv3.conv_in errRate = 0.12828534841537476 _00028_model.4.cv3.conv_out errRate = 0.11573690176010132 _00029_model.4.cv3.act_in errRate = 0.11573690176010132 _00029_model.4.cv3.act_out errRate = 0.11865927278995514 _00030_model.5.conv_in errRate = 0.11865927278995514 _00030_model.5.conv_out errRate = 0.12688465416431427 _00031_model.5.act_in errRate = 0.12688465416431427 _00031_model.5.act_out errRate = 0.13611656427383423 _00032_model.6.cv1.conv_in errRate = 0.13611656427383423 _00032_model.6.cv1.conv_out errRate = 0.08566196262836456 _00033_model.6.cv1.act_in errRate = 0.08566196262836456 _00033_model.6.cv1.act_out errRate = 0.08384081721305847 _00034_model.6.m.0.cv1.conv_in errRate = 0.08384081721305847 _00034_model.6.m.0.cv1.conv_out errRate = 0.12016722559928894 _00035_model.6.m.0.cv1.act_in errRate = 0.12016722559928894 _00035_model.6.m.0.cv1.act_out errRate = 0.13372856378555298 _00036_model.6.m.0.cv2.conv_in errRate = 0.13372856378555298 _00036_model.6.m.0.cv2.conv_out errRate = 0.08723128587007523 _00037_model.6.m.0.cv2.act_in errRate = 0.08723128587007523 _00037_model.6.m.0.cv2.act_out errRate = 0.0888165608048439 _00038_model.6.m.1.cv1.conv_in errRate = 0.0923202633857727 _00038_model.6.m.1.cv1.conv_out errRate = 0.10691310465335846 _00039_model.6.m.1.cv1.act_in errRate = 0.10691310465335846 _00039_model.6.m.1.cv1.act_out errRate = 0.11719808727502823 _00040_model.6.m.1.cv2.conv_in errRate = 0.11719808727502823 _00040_model.6.m.1.cv2.conv_out errRate = 0.12280922383069992 _00041_model.6.m.1.cv2.act_in errRate = 0.12280922383069992 _00041_model.6.m.1.cv2.act_out errRate = 0.13904298841953278 _00042_model.6.m.2.cv1.conv_in errRate = 0.11993970721960068 _00042_model.6.m.2.cv1.conv_out errRate = 0.11308033764362335 _00043_model.6.m.2.cv1.act_in errRate = 0.11308033764362335 _00043_model.6.m.2.cv1.act_out errRate = 0.11430805176496506 _00044_model.6.m.2.cv2.conv_in errRate = 0.11430805176496506 _00044_model.6.m.2.cv2.conv_out errRate = 0.13334615528583527 _00045_model.6.m.2.cv2.act_in errRate = 0.13334615528583527 _00045_model.6.m.2.cv2.act_out errRate = 0.14173951745033264 _00046_model.6.cv2.conv_in errRate = 0.13611656427383423 _00046_model.6.cv2.conv_out errRate = 0.15686963498592377 _00047_model.6.cv2.act_in errRate = 0.15686963498592377 _00047_model.6.cv2.act_out errRate = 0.163693368434906 _00048_model.6.cv3.conv_in errRate = 0.14533303678035736 _00048_model.6.cv3.conv_out errRate = 0.10818008333444595 _00049_model.6.cv3.act_in errRate = 0.10818008333444595 _00049_model.6.cv3.act_out errRate = 0.12623895704746246 _00050_model.7.conv_in errRate = 0.12623895704746246 _00050_model.7.conv_out errRate = 0.11763811856508255 _00051_model.7.act_in errRate = 0.11763811856508255 _00051_model.7.act_out errRate = 0.13873432576656342 _00052_model.8.cv1.conv_in errRate = 0.13873432576656342 _00052_model.8.cv1.conv_out errRate = 0.08493432402610779 _00053_model.8.cv1.act_in errRate = 0.08493432402610779 _00053_model.8.cv1.act_out errRate = 0.07582971453666687 _00054_model.8.m.0.cv1.conv_in errRate = 0.07582971453666687 _00054_model.8.m.0.cv1.conv_out errRate = 0.14061495661735535 _00055_model.8.m.0.cv1.act_in errRate = 0.14061495661735535 _00055_model.8.m.0.cv1.act_out errRate = 0.16039405763149261 _00056_model.8.m.0.cv2.conv_in errRate = 0.16039405763149261 _00056_model.8.m.0.cv2.conv_out errRate = 0.13419339060783386 _00057_model.8.m.0.cv2.act_in errRate = 0.13419339060783386 _00057_model.8.m.0.cv2.act_out errRate = 0.15217852592468262 _00058_model.8.cv2.conv_in errRate = 0.13873432576656342 _00058_model.8.cv2.conv_out errRate = 0.15616092085838318 _00059_model.8.cv2.act_in errRate = 0.15616092085838318 _00059_model.8.cv2.act_out errRate = 0.17604996263980865 _00060_model.8.cv3.conv_in errRate = 0.15473110973834991 _00060_model.8.cv3.conv_out errRate = 0.13436074554920197 _00061_model.8.cv3.act_in errRate = 0.13436074554920197 _00061_model.8.cv3.act_out errRate = 0.1602957397699356 _00062_model.9.cv1.conv_in errRate = 0.1602957397699356 _00062_model.9.cv1.conv_out errRate = 0.11659093201160431 _00063_model.9.cv1.act_in errRate = 0.11659093201160431 _00063_model.9.cv1.act_out errRate = 0.1178520917892456 _00064_model.9.m_in errRate = 0.1178520917892456 _00064_model.9.m_out errRate = 0.060698676854372025 _00065_model.9.m_in errRate = 0.060698676854372025 _00065_model.9.m_out errRate = 0.048983681946992874 _00066_model.9.m_in errRate = 0.048983681946992874 _00066_model.9.m_out errRate = 0.04338240996003151 _00067_model.9.cv2.conv_in errRate = 0.056752581149339676 _00067_model.9.cv2.conv_out errRate = 0.10899560153484344 _00068_model.9.cv2.act_in errRate = 0.10899560153484344 _00068_model.9.cv2.act_out errRate = 0.12165416032075882 _00069_model.10.conv_in errRate = 0.12165416032075882 _00069_model.10.conv_out errRate = 0.11829164624214172 _00070_model.10.act_in errRate = 0.11829164624214172 _00070_model.10.act_out errRate = 0.12541276216506958 _00071_model.11_in errRate = 0.12541276216506958 _00071_model.11_out errRate = 0.1254127323627472 _00072_model.12_in errRate = 0.1254127323627472 _00072_model.12_in errRate = 0.12623895704746246 _00072_model.12_out errRate = 0.12581397593021393 _00073_model.13.cv1.conv_in errRate = 0.12581397593021393 _00073_model.13.cv1.conv_out errRate = 0.0993901863694191 _00074_model.13.cv1.act_in errRate = 0.0993901863694191 _00074_model.13.cv1.act_out errRate = 0.09837651997804642 _00075_model.13.m.0.cv1.conv_in errRate = 0.09837651997804642 _00075_model.13.m.0.cv1.conv_out errRate = 0.10666487365961075 _00076_model.13.m.0.cv1.act_in errRate = 0.10666487365961075 _00076_model.13.m.0.cv1.act_out errRate = 0.10745485872030258 _00077_model.13.m.0.cv2.conv_in errRate = 0.10745485872030258 _00077_model.13.m.0.cv2.conv_out errRate = 0.115843266248703 _00078_model.13.m.0.cv2.act_in errRate = 0.115843266248703 _00078_model.13.m.0.cv2.act_out errRate = 0.11998312175273895 _00079_model.13.cv2.conv_in errRate = 0.12581397593021393 _00079_model.13.cv2.conv_out errRate = 0.12741047143936157 _00080_model.13.cv2.act_in errRate = 0.12741047143936157 _00080_model.13.cv2.act_out errRate = 0.12904389202594757 _00081_model.13.cv3.conv_in errRate = 0.12442249059677124 _00081_model.13.cv3.conv_out errRate = 0.13247622549533844 _00082_model.13.cv3.act_in errRate = 0.13247622549533844 _00082_model.13.cv3.act_out errRate = 0.1406361609697342 _00083_model.14.conv_in errRate = 0.1406361609697342 _00083_model.14.conv_out errRate = 0.12035928666591644 _00084_model.14.act_in errRate = 0.12035928666591644 _00084_model.14.act_out errRate = 0.11484494060277939 _00085_model.15_in errRate = 0.11484494060277939 _00085_model.15_out errRate = 0.11484494805335999 _00086_model.16_in errRate = 0.11484494805335999 _00086_model.16_in errRate = 0.11865927278995514 _00086_model.16_out errRate = 0.11659131199121475 _00087_model.17.cv1.conv_in errRate = 0.11659131199121475 _00087_model.17.cv1.conv_out errRate = 0.08819473534822464 _00088_model.17.cv1.act_in errRate = 0.08819473534822464 _00088_model.17.cv1.act_out errRate = 0.07866258174180984 _00089_model.17.m.0.cv1.conv_in errRate = 0.07866258174180984 _00089_model.17.m.0.cv1.conv_out errRate = 0.07065664231777191 _00090_model.17.m.0.cv1.act_in errRate = 0.07065664231777191 _00090_model.17.m.0.cv1.act_out errRate = 0.057614799588918686 _00091_model.17.m.0.cv2.conv_in errRate = 0.057614799588918686 _00091_model.17.m.0.cv2.conv_out errRate = 0.10141444206237793 _00092_model.17.m.0.cv2.act_in errRate = 0.10141444206237793 _00092_model.17.m.0.cv2.act_out errRate = 0.09200906753540039 _00093_model.17.cv2.conv_in errRate = 0.11659131199121475 _00093_model.17.cv2.conv_out errRate = 0.10760253667831421 _00094_model.17.cv2.act_in errRate = 0.10760253667831421 _00094_model.17.cv2.act_out errRate = 0.11112961918115616 _00095_model.17.cv3.conv_in errRate = 0.10040128231048584 _00095_model.17.cv3.conv_out errRate = 0.1444503366947174 _00096_model.17.cv3.act_in errRate = 0.1444503366947174 _00096_model.17.cv3.act_out errRate = 0.138624370098114 _00097_model.18.conv_in errRate = 0.138624370098114 _00097_model.18.conv_out errRate = 0.15020611882209778 _00098_model.18.act_in errRate = 0.15020611882209778 _00098_model.18.act_out errRate = 0.15838798880577087 _00099_model.19_in errRate = 0.15838798880577087 _00099_model.19_in errRate = 0.11484494060277939 _00099_model.19_out errRate = 0.13492365181446075 _00100_model.20.cv1.conv_in errRate = 0.13492365181446075 _00100_model.20.cv1.conv_out errRate = 0.1408357173204422 _00101_model.20.cv1.act_in errRate = 0.1408357173204422 _00101_model.20.cv1.act_out errRate = 0.1426309496164322 _00102_model.20.m.0.cv1.conv_in errRate = 0.1426309496164322 _00102_model.20.m.0.cv1.conv_out errRate = 0.12285890430212021 _00103_model.20.m.0.cv1.act_in errRate = 0.12285890430212021 _00103_model.20.m.0.cv1.act_out errRate = 0.12916986644268036 _00104_model.20.m.0.cv2.conv_in errRate = 0.12916986644268036 _00104_model.20.m.0.cv2.conv_out errRate = 0.13812172412872314 _00105_model.20.m.0.cv2.act_in errRate = 0.13812172412872314 _00105_model.20.m.0.cv2.act_out errRate = 0.14069794118404388 _00106_model.20.cv2.conv_in errRate = 0.13492365181446075 _00106_model.20.cv2.conv_out errRate = 0.15035104751586914 _00107_model.20.cv2.act_in errRate = 0.15035104751586914 _00107_model.20.cv2.act_out errRate = 0.1491565853357315 _00108_model.20.cv3.conv_in errRate = 0.14448954164981842 _00108_model.20.cv3.conv_out errRate = 0.15475969016551971 _00109_model.20.cv3.act_in errRate = 0.15475969016551971 _00109_model.20.cv3.act_out errRate = 0.15354607999324799 _00110_model.21.conv_in errRate = 0.15354607999324799 _00110_model.21.conv_out errRate = 0.1348518282175064 _00111_model.21.act_in errRate = 0.1348518282175064 _00111_model.21.act_out errRate = 0.1416780948638916 _00112_model.22_in errRate = 0.1416780948638916 _00112_model.22_in errRate = 0.12541276216506958 _00112_model.22_out errRate = 0.133895605802536 _00113_model.23.cv1.conv_in errRate = 0.133895605802536 _00113_model.23.cv1.conv_out errRate = 0.12326427549123764 _00114_model.23.cv1.act_in errRate = 0.12326427549123764 _00114_model.23.cv1.act_out errRate = 0.13576211035251617 _00115_model.23.m.0.cv1.conv_in errRate = 0.13576211035251617 _00115_model.23.m.0.cv1.conv_out errRate = 0.13252092897891998 _00116_model.23.m.0.cv1.act_in errRate = 0.13252092897891998 _00116_model.23.m.0.cv1.act_out errRate = 0.14680726826190948 _00117_model.23.m.0.cv2.conv_in errRate = 0.14680726826190948 _00117_model.23.m.0.cv2.conv_out errRate = 0.12699651718139648 _00118_model.23.m.0.cv2.act_in errRate = 0.12699651718139648 _00118_model.23.m.0.cv2.act_out errRate = 0.13754235208034515 _00119_model.23.cv2.conv_in errRate = 0.133895605802536 _00119_model.23.cv2.conv_out errRate = 0.13765279948711395 _00120_model.23.cv2.act_in errRate = 0.13765279948711395 _00120_model.23.cv2.act_out errRate = 0.14234159886837006 _00121_model.23.cv3.conv_in errRate = 0.13985086977481842 _00121_model.23.cv3.conv_out errRate = 0.15157805383205414 _00122_model.23.cv3.act_in errRate = 0.15157805383205414 _00122_model.23.cv3.act_out errRate = 0.14827151596546173 _00123_model.24.m.0_in errRate = 0.138624370098114 ******************* _00123_model.24.m.0_out errRate = 0.02405344322323799 ******************* _00124_model.24.m.1_in errRate = 0.15354607999324799 ******************* _00124_model.24.m.1_out errRate = 0.021177854388952255 ******************* _00125_model.24.m.2_in errRate = 0.14827151596546173 ******************* _00125_model.24.m.2_out errRate = 0.018881356343626976 ******************* 我已经在误差变化较大的地方做了标注 又几个疑问 1.我理解的误差是逐层累积越来越大的,对吗? 2.网络24层的误差在输入时0.13 但是输出误差却突然变小了,这是怎么回事呢 3.虽然最后一层误差变小,但是在结果上来看,并没又提现出来,误差还是很大 10%左右 4.为什么在第2层使用cov1 误差突然就翻倍了?这个算子在其他地方用也没有发现误差很大呀 5.我该怎么解决呢
热门帖子
精华帖子