模型转换报错

已解决

Darth Nihilus2026-05-15

243

yolov5n.yaml文件如下：

model_parameters:

onnx_model: '/data/horizon_j5/data/yolov5n.onnx'

march: "bayes"

layer_out_dump: False

working_dir: '../model_output'

output_model_file_prefix: 'yolov5n-det'

remove_node_type: "Dequantize"

node_info: {"/model.10/m/m.0/attn/Softmax": {'ON': 'BPU','InputType': 'int8','OutputType': 'int8'}}

input_parameters:

input_name: 'images'

input_shape: '1x3x640x640'

input_type_rt: 'nv12'

input_type_train: 'rgb'

input_layout_train: 'NCHW'

mean_value: '0.0 0.0 0.0'

scale_value: '0.00392157 0.00392157 0.00392157'

calibration_parameters:

cal_dir: '/data/horizon_j5/data/calib_data'

compiler_parameters:

compile_mode: 'latency'

core_num: 1

optimize_level: 'O2'

jobs: 8

运行与报错如下：

hb_mapper makertbin --config /data/horizon_j5/data/yolov5n.yaml --model-type onnx

/usr/local/lib/python3.8/dist-packages/paramiko/pkey.py:100: CryptographyDeprecationWarning: TripleDES has been moved to cryptography.hazmat.decrepit.ciphers.algorithms.TripleDES and will be removed from this module in 48.0.0.

"cipher": algorithms.TripleDES,

/usr/local/lib/python3.8/dist-packages/paramiko/transport.py:259: CryptographyDeprecationWarning: TripleDES has been moved to cryptography.hazmat.decrepit.ciphers.algorithms.TripleDES and will be removed from this module in 48.0.0.

"class": algorithms.TripleDES,

2026-05-15 09:49:15,773 INFO log will be stored in //hb_mapper_makertbin.log

2026-05-15 09:49:15,774 INFO Start hb_mapper....

2026-05-15 09:49:15,774 INFO hbdk version 3.49.13

2026-05-15 09:49:15,774 INFO horizon_nn version 1.0.9

2026-05-15 09:49:15,774 INFO hb_mapper version 1.24.1

2026-05-15 09:49:15,774 INFO Start Model Convert....

2026-05-15 09:49:15,784 INFO Using onnx model file: /data/horizon_j5/data/yolov5n.onnx

2026-05-15 09:49:15,804 INFO Model has 1 inputs according to model file

2026-05-15 09:49:15,805 INFO nv12 input type rt received.

2026-05-15 09:49:15,806 ERROR Parse 'calibration_parameters' failed!

2026-05-15 09:49:15,806 ERROR yaml file parse failed. Please double check your config file inputs

2026-05-15 09:49:15,808 ERROR Wrong cal_dir num given, cal_dir num 0 is not equal to input num 1

算法工具链

社区活动征程5

0/600

DR_KAN
Lv.5
你的/data/horizon_j5/data/calib_data下面直接放校准数据就行，比如npy或者bin文件
2026-05-15
0
3
- Darth Nihilus回复DR_KAN:
  我重新修改了地址，dir对了但是报错是：
  2026-05-17 16:06:39,944 WARNING The calibration dir name suffix is not the same as the value float32 of the parameter cal_data_type, the parameter setting will prevail
  2026-05-17 16:06:39,945 INFO custom_op does not exist, skipped
  2026-05-17 16:06:39,945 WARNING Input node images's input_source not set, it will be set to pyramid by default
  2026-05-17 16:06:39,948 INFO *******************************************
  2026-05-17 16:06:39,948 INFO First calibration picture name: test.npy
  2026-05-17 16:06:39,949 INFO First calibration picture md5:
  d7193da8cb6f902d097599c5a6bc67da /data/horizon_j5/data/calib_data/test.npy
  2026-05-17 16:06:39,965 INFO *******************************************
  2026-05-17 16:06:39,968 ERROR load cal data for input 'images' error
  2026-05-17 16:06:39,970 ERROR cannot reshape array of size 1228832 into shape (1,3,640,640)
  2026-05-17
  0
  回复
- DR_KAN回复Darth Nihilus:
  你这个npy有head数据，不纯净，建议转成bin再做校准吧，或者把head数据删了。已经快成功了。
  2026-05-18
  0
  回复
- Darth Nihilus回复DR_KAN:
  老师，我这边将.bin拷贝至板中，根据教程：
  ping 192.168.0.10
  PING 192.168.0.10 (192.168.0.10) 56(84) bytes of data.
  64 bytes from 192.168.0.10: icmp_seq=1 ttl=63 time=0.308 ms
  64 bytes from 192.168.0.10: icmp_seq=2 ttl=63 time=0.415 ms
  64 bytes from 192.168.0.10: icmp_seq=3 ttl=63 time=0.315 ms
  64 bytes from 192.168.0.10: icmp_seq=4 ttl=63 time=0.323 ms
  --- 192.168.0.10 ping statistics ---
  4 packets transmitted, 4 received, 0% packet loss, time 3072ms
  rtt min/avg/max/mdev = 0.308/0.340/0.415/0.043 ms
  scp model_output/yolov5n-det.bin root@192.168.0.10:/userdata
  ssh: connect to host 192.168.0.10 port 22: Connection refused
  lost connection
  这个如何解决呢？
  2026-05-19
  0
  回复
YCJ
Lv.4
这是没有检测到校准集，你的cal_data_dir参数写错了吧
2026-05-15
0
7
- Darth Nihilus回复YCJ:
  我重新修改了地址，dir对了但是报错是：
  2026-05-17 16:06:39,944 WARNING The calibration dir name suffix is not the same as the value float32 of the parameter cal_data_type, the parameter setting will prevail
  2026-05-17 16:06:39,945 INFO custom_op does not exist, skipped
  2026-05-17 16:06:39,945 WARNING Input node images's input_source not set, it will be set to pyramid by default
  2026-05-17 16:06:39,948 INFO *******************************************
  2026-05-17 16:06:39,948 INFO First calibration picture name: test.npy
  2026-05-17 16:06:39,949 INFO First calibration picture md5:
  d7193da8cb6f902d097599c5a6bc67da /data/horizon_j5/data/calib_data/test.npy
  2026-05-17 16:06:39,965 INFO *******************************************
  2026-05-17 16:06:39,968 ERROR load cal data for input 'images' error
  2026-05-17 16:06:39,970 ERROR cannot reshape array of size 1228832 into shape (1,3,640,640)
  2026-05-17
  0
  回复
- YCJ回复Darth Nihilus:
  校准集的处理有问题，你的输入是1*3*640*640=1228800，现在你的npy文件多了32个元素，检查一下校准集的处理是否正确，要与模型前处理一致
  2026-05-18
  0
  回复
- Darth Nihilus回复YCJ:
  老师，我这边将.bin拷贝至板中，根据教程：
  ping 192.168.0.10
  PING 192.168.0.10 (192.168.0.10) 56(84) bytes of data.
  64 bytes from 192.168.0.10: icmp_seq=1 ttl=63 time=0.308 ms
  64 bytes from 192.168.0.10: icmp_seq=2 ttl=63 time=0.415 ms
  64 bytes from 192.168.0.10: icmp_seq=3 ttl=63 time=0.315 ms
  64 bytes from 192.168.0.10: icmp_seq=4 ttl=63 time=0.323 ms
  --- 192.168.0.10 ping statistics ---
  4 packets transmitted, 4 received, 0% packet loss, time 3072ms
  rtt min/avg/max/mdev = 0.308/0.340/0.415/0.043 ms
  scp model_output/yolov5n-det.bin root@192.168.0.10:/userdata
  ssh: connect to host 192.168.0.10 port 22: Connection refused
  lost connection
  这个如何解决呢？
  2026-05-19
  0
  回复
- YCJ回复Darth Nihilus:
  检查一下板端有没有ssh服务，是否启动
  2026-05-19
  0
  回复
- Darth Nihilus回复YCJ:
  这个如何检查啊？我看说是在开发机上ddk目录下，但是我docker环境中没有找到这个目录，想请教下~
  2026-05-19
  0
  回复
- YCJ回复Darth Nihilus:
  你ssh能连板子吗？不能连的话在板子上装ssh服务呀，这个跟docker没有关系
  2026-05-19
  0
  回复
- Darth Nihilus回复YCJ:
  我就是不知道怎么在板子上安ssh。我看教程说是在开发机上运行.sh脚本远程安装
  2026-05-19
  0
  回复