2种模型的预处理相同，输入到om的数据和输入到onnx的数据完全一致，这里对比的直接是model输出的结果，不存在任何后处理操作。

ATC模型转换命令为:atc --model=/home/projects/facefusion-onnxrun-main/weights/yoloface_8n.onnx --framework=5 --output=yoloface_8n --input_format=NCHW --input_shape="images:1,3,640,640" --enable_small_channel=1 --log=error --soc_version=Ascend310P3

onnx model输出：[[[1.1692884e+01 1.8548105e+01 2.3026794e+01 ... 5.7547491e+02 

   6.0357910e+02 6.1490625e+02] 

  [7.2354121e+00 4.8529878e+00 4.7415161e+00 ... 5.5844067e+02 

   5.4418768e+02 5.4982312e+02] 

  [2.3492554e+01 3.6706470e+01 4.9392159e+01 ... 1.2872379e+02 

   7.3143799e+01 9.7037231e+01] 

  ... 

  [5.3061609e+00 9.7035160e+00 1.4538417e+01 ... 5.9863354e+02 

   6.2501697e+02 6.2823633e+02] 

  [4.9153900e+00 5.4273720e+00 7.3639407e+00 ... 6.2493213e+02 

   6.1854565e+02 6.2337463e+02] 

  [2.1815300e-02 1.4590591e-02 2.3108929e-02 ... 4.2995811e-03 

   2.6901066e-03 5.9473515e-02]]]

OM模型输出：

[[[8.34375000e+00 1.50781250e+01 1.83125000e+01 ... 5.26000000e+02 

   5.50000000e+02 5.95500000e+02] 

  [1.45468750e+01 4.42968750e+00 3.02148438e+00 ... 6.13000000e+02 

   6.15500000e+02 6.13000000e+02] 

  [1.65000000e+01 1.64062500e+01 1.16718750e+01 ... 2.28000000e+02 

   1.81500000e+02 1.98000000e+02] 

  ... 

  [9.07812500e+00 1.92031250e+01 2.13125000e+01 ... 5.75500000e+02 

   5.93000000e+02 6.18500000e+02] 

  [1.30546875e+01 1.06093750e+01 5.92187500e+00 ... 6.22500000e+02 

   6.20500000e+02 6.25000000e+02] 

  [1.99584961e-01 2.10449219e-01 2.36206055e-01 ... 3.78036499e-03 

   6.85119629e-03 1.10839844e-01]]]