Optimizing the operation of neural analytics on GPU in Linux OS

On the page:

General information

It can take several minutes to launch neural analytics algorithms on Nvidia GPU after server restart. Meanwhile, the neural models are optimized for the current GPU type.

You can use the caching function to ensure that this operation is performed only once. Caching saves the optimization results on the hard drive and uses it for the subsequent analytics runs.

Starting with DetectorPack 3.9, a utility was added to the Neuro Pack add-ons (see Installing DetectorPack add-ons), which allows you to create GPU neural network caches without using Axxon One. The presence of the cache speeds up the initialization and optimizes video memory consumption.

Optimizing the operation of neural analytics on GPU

To optimize the operation of the neural analytics on GPU, do the following:

Stop the server (see Starting and stopping the Axxon One Server in Linux OS).
Attention!
If the system has the software running on GPU, it is necessary to stop its operation.
Login as ngp superuser:
1. In the command prompt, run the command:
```
sudo su ngp
```
2. Enter the password for the root superuser.
Create a folder with a custom name to store the cache. For example:
```
mkdir /opt/AxxonSoft/AxxonOne/gpucache
```

Change folder permissions:

chmod -R 777 /opt/AxxonSoft/AxxonOne/gpucache

Create the GPU_CACHE_DIR system variable:
1. Go to the /opt/AxxonSoft/AxxonOne/ folder:
```
cd /opt/AxxonSoft/AxxonOne
```
2. Open the instance.conf file for editing:
```
nano instance.conf
```
3. Add the following line to the file:
```
export GPU_CACHE_DIR="/opt/AxxonSoft/AxxonOne/gpucache"
```
  Attention!
  If you change the server configuration (see Changing the configuration of the Axxon One Server in Linux OS) or update to a new version of Axxon One, the system variables previously added to the instance.conf file will be deleted (see Creating system variables in Linux OS).
4. Save the file using the Ctrl+O keyboard shortcut.
5. Exit file editing mode using the Ctrl+X keyboard shortcut.
6. Run the following command in the terminal:
```
export GPU_CACHE_DIR="/opt/AxxonSoft/AxxonOne/gpucache"
```
Go to the /opt/AxxonSoft/DetectorPack/ folder:
```
cd /opt/AxxonSoft/DetectorPack
```
Run the following command:
```
./NeuroPackGpuCacheGenerator
```
Attention!
If more than one Nvidia GPU is available, you will be able to select one. To do this, specify a number from 0 to 3 which corresponds to the required device in the list.

Optimizing the operation of the neural analytics on GPU is complete. The utility will create the caches of four neural networks included in the Neuro Pack add-ons:

GeneralNMHuman_v1.0GPU_onnx.ann—person;
smokeScanned_v1_onnx.ann—smoke detection;
fireScanned_v1_onnx.ann—fire detection;
reid_15_0_256__osnetfpn_segmentation_noise_20_common_29_onnx.ann—search for the similar in the Neural tracker (see Similitude search).

Creating GPU neural network caches using parameters

-p is a parameter to create a cache for a particular neural network.
Command example:

./NeuroPackGpuCacheGenerator -p /opt/AxxonSoft/DetectorPack/NeuroSDK/GeneralNMHumanAndVehicle_Nano_v1.0_GPU_onnx.ann

-v is a parameter to output the procedure log to the console during cache generation.
Command example to automatically create caches of four neural networks included in the Neuro Pack add-ons with log output:
```
./NeuroPackGpuCacheGenerator -v
```
--int8=1 is a parameter to create a quantized version of the cache for those neural networks for which quantization is available. By default, the --int8=0 parameter is disabled.
Command example:
```
./NeuroPackGpuCacheGenerator -p /opt/AxxonSoft/DetectorPack/NeuroSDK/GeneralNMHumanAndVehicle_Nano_v1.0_GPU_onnx.ann --int8=1
```
Attention!
The neural networks for which the quantization mode is available are included in the Neuro Pack add-ons together with the *.info file.

The neural networks for which the quantization mode is available (see Neural tracker, Stopped object detector, Neural counter):

GeneralNMCar_v1.0GPU_onnx.ann—Vehicle.
GeneralNMHuman_v1.0GPU_onnx.ann—Person.
GeneralNMHumanTopView_v0.8GPU_onnx.ann—Person (top-down view).

Starting with DetectorPack 3.11, the following neural networks were added:

GeneralNMHumanAndVehicle_Nano_v1.0_GPU_onnx.ann—Person and vehicle (Nano).
GeneralNMHumanAndVehicle_Medium_v1.0_GPU_onnx.ann—Person and vehicle (Medium).
GeneralNMHumanAndVehicle_Large_v1.0_GPU_onnx.ann—Person and vehicle (Large).

Starting with DetectorPack 3.12, the following neural networks were added:

GeneralNMHumanTopView_Nano_v1.0_GPU_onnx.ann—Person (top-down view Nano).
GeneralNMHumanTopView_Medium_v1.0_GPU_onnx.ann—Person (top-down view Medium).
GeneralNMHumanTopView_Large_v1.0_GPU_onnx.ann—Person (top-down view Large).

Page tree

Documentation for Axxon One 2.0. Documentation for other versions of Axxon One is available too.

General information

Optimizing the operation of neural analytics on GPU

Creating GPU neural network caches using parameters