OS version : VIM3L_Ubuntu-server-bionic_Linux-4.9_arm64_EMMC_V20191231.7z
The package has been upgraded to the latest.
Use acuity-toolkit 5.0.0 convert inception_v3 success, But when run demo in board got NBG fail error:
khadas@Khadas:~/c_code$ lsmod | grep galcore
galcore 331776 0
khadas@Khadas:~/c_code$ ./inceptionv3 inception_v3.nb goldfish_299x299.jpg
D [setup_node:367]Setup node id[0] uid[0] op[NBG]
D [print_tensor:129]in : id[ 1] shape[ 3, 299, 299, 1 ] fmt[u8 ] qnt[ASM zp=137, scale=0.007292]
D [print_tensor:129]out: id[ 0] shape[ 1001, 1 ] fmt[f16] qnt[NONE]
D [optimize_node:311]Backward optimize neural network
D [optimize_node:318]Forward optimize neural network
I [compute_node:260]Create vx node
E [compute_node:286]Create node[0] NBG fail
E [vnn_CreateNeuralNetwork:174]CHECK PTR 174
E [main:210]CHECK PTR 210
Frank
#2
@jujuede Did you use the model demo which was trained by yourself ?
No, I followed this tutorial.
I found that in the official pre-compiled demo, there are two versions of the inception_v3 demo.
khadas@Khadas:~/aml_npu_demo_binaries/inceptionv3$ ls
VIM3 VIM3L dog_299x299.jpg goldfish_299x299.jpg imagenet_slim_labels.txt
The size of the model file used by the two versions of the demo is not the same:
khadas@Khadas:~/aml_npu_demo_binaries/inceptionv3$ ll -h VIM3L/inception_v3.nb VIM3/inception_v3.nb
-rw-rw-r-- 1 khadas khadas 28M Mar 8 03:51 VIM3/inception_v3.nb
-rw-rw-r-- 1 khadas khadas 20M Mar 8 03:51 VIM3L/inception_v3.nb
I used the VIM3 executable to execute the VIM3L model file with success.
khadas@Khadas:~/aml_npu_demo_binaries/inceptionv3$ export VSI_NN_LOG_LEVEL=0 && ./VIM3/inceptionv3 ./VIM3L/inception_v3.nb goldfish_299x299.jpg
Create Neural Network: 33ms or 33699us
Verify...
Verify Graph: 2ms or 2099us
Start run graph [1] times...
Run the 1 time: 79ms or 79635us
vxProcessGraph execution time:
Total 79ms or 79689us
Average 79.69ms or 79689.00us
--- Top5 ---
2: 0.826660
795: 0.008255
974: 0.004066
393: 0.002390
408: 0.002285
Executing a VIM3 model file using the VIM3 executable file fails.
khadas@Khadas:~/aml_npu_demo_binaries/inceptionv3$ ./VIM3/inceptionv3 ./VIM3/inception_v3.nb goldfish_299x299.jpg
E [compute_node:286]Create node[0] NBG fail
E [vnn_CreateNeuralNetwork:174]CHECK PTR 174
E [main:210]CHECK PTR 210
Is the model data conversion command of VIM3L version different from VIM3 version? I did not find a specific tutorial for the VIM3L demo in this repo
Frank
#4
@jujuede the conversion no difference . Without replacement, can the original demo be used ?
The size of the nb file I converted according to the tutorial is the same as the nb file size in VIM3 demo:
(vim3l) ➜ nbg_unify_inception_v3 ls -al inception_v3.nb
-rw-r--r-- 1 zqh zqh 28807168 3月 6 18:27 inception_v3.nb
khadas@Khadas:~/aml_npu_demo_binaries/inceptionv3$ ll VIM3/inception_v3.nb
-rw-rw-r-- 1 khadas khadas 28807168 Mar 8 03:51 VIM3/inception_v3.nb
NBG fail error occurs when using VIM3 version .nb file:
khadas@Khadas:~/aml_npu_demo_binaries/inceptionv3/VIM3$ ./inceptionv3 inception_v3.nb ../goldfish_299x299.jpg
E [compute_node:286]Create node[0] NBG fail
E [vnn_CreateNeuralNetwork:174]CHECK PTR 174
E [main:210]CHECK PTR 210
The VIM3L version .nb file size is:
khadas@Khadas:~/aml_npu_demo_binaries/inceptionv3/VIM3L$ ls -al inception_v3.nb
-rw-rw-r-- 1 khadas khadas 20799872 Mar 8 03:51 inception_v3.nb
Frank
#6
@jujuede If you want to use in VIM3L , the parameter optimize in the script which name 2_export_case_code , you shuld set it to be VIPNANOQI_PID0X99
1 Like