专栏算法工具链使用ROI Infer前推报错

使用ROI Infer前推报错

已解决
白菜飘落2024-02-06
157
8

芯片型号：J5

问题定位：板端部署

问题描述：使用hbDNNRoiInfer接口进行推理时，接口报错。报错信息如下：

[03:46:38:435:266] pid=20606 tid=20619 [HBRT 3.14.25.0] (line 3607 in hbrtInternalCanonicalizeInputInfos from file=f3a31eef122a41bb11cce71188f99333fc1ffdf9) input image stride = 128 is smaller than image width = 1280. May cause undefined behavior!

[03:46:38:435:493] pid=20606 tid=20619 [HBRT 3.14.25.0] (line 74 in hbrtInternalGetRoiInputInfo from file=d40cb715974a0360a49d287a19a2c0f4f19891c6) resizer config is invalid. image h/w/s = 640,1280,128

hbrtErrorInvalidConfig 3.14.25 file=d40cb715974a0360a49d287a19a2c0f4f19891c6 179

hbrtErrorInvalidConfig 3.14.25 file=d40cb715974a0360a49d287a19a2c0f4f19891c6 246

hbrtErrorInvalidConfig 3.14.25 file=f3a31eef122a41bb11cce71188f99333fc1ffdf9 3001

hbrtErrorInvalidConfig 3.14.25 file=f3a31eef122a41bb11cce71188f99333fc1ffdf9 4142

hbrtErrorInvalidConfig 3.14.25 file=f3a31eef122a41bb11cce71188f99333fc1ffdf9 4258

hbrtErrorInvalidConfig 3.14.25 file=f3a31eef122a41bb11cce71188f99333fc1ffdf9 4412

hbrtErrorInvalidConfig 3.14.25 file=f3a31eef122a41bb11cce71188f99333fc1ffdf9 4482

hbrtErrorInvalidConfig 3.14.25 file=f3a31eef122a41bb11cce71188f99333fc1ffdf9 3668

hbrtErrorInvalidConfig 3.14.25 file=f3a31eef122a41bb11cce71188f99333fc1ffdf9 3964

hbrtErrorInvalidConfig 3.14.25 file=f3a31eef122a41bb11cce71188f99333fc1ffdf9 4038

 [E][DNN][hbm_exec_plan.cpp:1030](503198436) [HBRT ERROR]hbrtErrorInvalidConfig 

 [E][DNN][multi_model_task.cpp:1302](503198436) ri Start failed, ri_id: 2

算法工具链
征程5
评论3
0/1000
  • Huanghui
    Lv.5

    你好，从日志看是输入数据的shape信息与模型所需的不符，有两种处理方式：1）自己验证，通过hrt_model_exec model_info查看模型所需数据，然后推理之前将input data处理成模型所需要的大小。2）提供测试模型和测试程序，我们在服务器上验证问题。

    2024-02-18
    0
    5
    • 白菜飘落回复Huanghui:

      你好，百度云盘的链接【超级会员V9】通过百度网盘分享的文件：to-horiz....xz 链接：https://pan.baidu.com/s/1X9qgxnzdJ_R_ORb5FLUdrA?pwd=2d6a 提取码：2d6a 复制这段内容打开「百度网盘APP 即可获取」

      2024-02-19
      0
    • 四木小子回复Huanghui:

      同问，关注一下

      2024-02-23
      0
    • 白菜飘落回复Huanghui:

      你好，请问有什么进展吗？

      2024-02-29
      0
    • Huanghui回复Huanghui:

      该问题是hbDNNTensor使用错误所致，请按照如下截图中的代码修改

      2024-03-06
      0
    • Huanghui回复Huanghui:

      for (auto &oneInputTensor : inputTensors) { oneInputTensor.sysMem[0] = inputImgLumaMem; oneInputTensor.sysMem[1] = inputImgChromaMem; oneInputTensor.properties.validShape.dimensionSize[2] = imgHeight; oneInputTensor.properties.validShape.dimensionSize[3] = imgWidth; oneInputTensor.properties.alignedShape = oneInputTensor.properties.validShape; }

      2024-03-06
      0
  • Huanghui
    Lv.5
    2024-03-06
    0
    0
  • 白菜飘落
    Lv.1

    好的，非常感谢。我改了之后就好了。

    2024-03-11
    0
    0