1. 问题现象
输入图片都是640x640格式时，大部分图推理结果torch和acl推理结果基本一致

比如：

acl 结果: [tensor([[132.6875, 125.6250, 321.7500, 563.2500,   0.9279,   0.0000]])]
torch 结果: [tensor([[132.5750, 127.8934, 321.8381, 566.1253,   0.9390,   0.0000]])]

[tensor([[156.3125, 133.1250, 274.7500, 528.1250,   0.9292,   0.0000]])]
[tensor([[155.1877, 123.3861, 274.8097, 528.6846,   0.9393,   0.0000]])]

[tensor([[125.5000, 127.5000, 319.9375, 564.2500,   0.9260,   0.0000]])]
[tensor([[125.7130, 128.4243, 320.4398, 568.8907,   0.9316,   0.0000]])]

[tensor([[335.1875, 219.9375, 389.6406, 411.1250,   0.9513,   0.0000]])]
[tensor([[335.0974, 219.3627, 389.7469, 409.2925,   0.9547,   0.0000]])]

[tensor([[366.9375, 239.9375, 422.4844, 389.0000,   0.9493,   0.0000]])]
[tensor([[366.6733, 239.5600, 422.7210, 389.5173,   0.9521,   0.0000]])]

少部分在推理时torch推理存在结果，但是acl推理后不出结果,  有些图片在旋转后acl推理又可以出结果

2. 软件版本

torch                  2.4.0

python              3.9.9

cann                  8.0.RC3.alpha002

openEuler      22.03LTS

+--------------------------------------------------------------------------------------------------------+

| npu-smi 24.1.rc3                                 Version: 24.1.rc3                                     |

+-------------------------------+-----------------+------------------------------------------------------+

| NPU     Name                  | Health          | Power(W)     Temp(C)           Hugepages-Usage(page) |

| Chip    Device                | Bus-Id          | AICore(%)    Memory-Usage(MB)                        |

+===============================+=================+======================================================+

| 1       310P3                 | OK              | NA           70                0     / 0             |

| 0       0                     | 0000:06:00.0    | 0            1603 / 44280                            |

+-------------------------------+-----------------+------------------------------------------------------+

| 1       310P3                 | OK              | NA           67                0     / 0             |

| 1       1                     | 0000:06:00.0    | 0            1313 / 43693                            |

+===============================+=================+======================================================+

| 3       310P3                 | OK              | NA           64                0     / 0             |

| 0       2                     | 0000:07:00.0    | 0            1586 / 44280                            |

+-------------------------------+-----------------+------------------------------------------------------+

| 3       310P3                 | OK              | NA           64                0     / 0             |

| 1       3                     | 0000:07:00.0    | 0            1328 / 43693                            |

+===============================+=================+======================================================+

+-------------------------------+-----------------+------------------------------------------------------+

有什么方法可以分析下是哪里的问题么