8. Advanced Guide to Model Deployment
8.1. Overview
This chapter introduces how to use the NPU-related sample programs on the development board. For the source code of the relevant sample programs, refer to the msp/sample/npu directory in the SDK. For how to compile the NPU sample code, please refer to the “AX SDK User Guide”.
8.1.1. Running the Example
Running Preparation
For AX650A, AX650N, M76H, AX630C development boards, NPU related examples have been pre-installed in the /opt/bin/ path, namely sample_npu_classification and sample_npu_yolov5s.
For the AX620Q development board, since the 16M NorFlash solution is adopted by default, the above two examples are not included in the file system. You can mount the msp/out/bin/ path in the SDK to the file system of the development board through NFS network mounting to obtain the above examples.
If it is prompted that the board space is insufficient, you can solve it by mounting the folder.
Example of mounting ARM development board on MacOS
Hint
Due to the limited space on the board, folder sharing is usually required during testing. At this time, it is necessary to share the ARM development board with the host. Here we only take MacOS as an example.
The development machine needs the NFS service to mount the ARM development board, and the MacOS system comes with the NFS service. You only need to create the /etc/exports folder, and nfsd will automatically start and start using exports.
/etc/exports can be configured as follows:
/path/your/sharing/directory -alldirs -maproot=root:wheel -rw -network xxx.xxx.xxx.xxx -mask 255.255.255.0
Parameter Explanation
Parameter name |
Meaning |
|---|---|
alldirs |
Share all files under the |
network |
Mount the ARM development board IP address, which can be a network segment address. |
mask |
Subnet mask, usually 255.255.255.0. |
maproot |
Mapping rule. When |
rw |
Read and write operations, enabled by default |
Modifying /etc/exports requires restarting the nfsd service
sudo nfsd restart
If the configuration is successful, you can use
sudo showmount -e
Command to view the mount information, for example, the output is /Users/skylake/board_nfs 10.168.21.xx. After configuring the development machine, you need to execute the mount command on the ARM side.
mount -t nfs -o nolock,tcp macos_ip:/your/shared/directory /mnt/directory
If there is a permission problem, you need to check whether the maproot parameter is correct.
Hint
The network parameter can be configured in the form of a network segment, such as: 10.168.21.0. If Permission denied appears when mounting a single IP, you can try mounting within the network segment.
Classification Model
The following print information is based on the output of the AX650N development board. The print information of non-AX650N development boards shall be subject to the actual print.
/root # sample_npu_classification -m /opt/data/npu/models/mobilenetv2.axmodel -i /opt/data/npu/images/cat.jpg -r 10
--------------------------------------
model file : /opt/data/npu/models/mobilenetv2.axmodel
image file : /opt/data/npu/images/cat.jpg
img_h, img_w : 224 224
--------------------------------------
Engine creating handle is done.
Engine creating context is done.
Engine get io info is done.
Engine alloc io is done.
Engine push input is done.
--------------------------------------
topk cost time:0.07 ms
9.5094, 285
9.3773, 282
9.2452, 281
8.5849, 283
7.6603, 287
--------------------------------------
Repeat 10 times, avg time 0.72 ms, max_time 0.72 ms, min_time 0.72 ms
--------------------------------------
Detection Model
/root # sample_npu_yolov5s -m /opt/data/npu/models/yolov5s.axmodel -i /opt/data/npu/images/dog.jpg -r 10
--------------------------------------
model file : /opt/data/npu/models/yolov5s.axmodel
image file : /opt/data/npu/images/dog.jpg
img_h, img_w : 640 640
--------------------------------------
Engine creating handle is done.
Engine creating context is done.
Engine get io info is done.
Engine alloc io is done.
Engine push input is done.
--------------------------------------
post process cost time:2.25 ms
--------------------------------------
Repeat 10 times, avg time 7.65 ms, max_time 7.66 ms, min_time 7.65 ms
--------------------------------------
detection num: 3
16: 91%, [ 138, 218, 310, 541], dog
2: 69%, [ 470, 76, 690, 173], car
1: 56%, [ 158, 120, 569, 420], bicycle
--------------------------------------
8.2. Other Examples
Please refer to our open source projects on github: