Pre-trained Free AI Application Libraries for RZ/V2L: Difference between revisions
(→Sample Videos: Added Inference time) |
|||
(6 intermediate revisions by the same user not shown) | |||
Line 10: | Line 10: | ||
** Face recognition and spoof detection (Work in progress | ** Face recognition and spoof detection (Work in progress | ||
[https://www.renesas.com/us/en/document/lbr/renesas-rzv-pre-trained-ai-library '''Pre-Trained AI Article'''] | [https://www.renesas.com/us/en/document/lbr/renesas-rzv-pre-trained-ai-library '''Pre-Trained AI Article'''] | ||
= Addition Notes = | |||
The Pre-trained models include pre-compiled applications as well as AI Models translated to run on the DRP-AI hardware. These files are located in the "exe" folder for each Pre-trained Application. These precompiled application are compiled for the Renesas RZV2L EVK using the Coral MIPI Camera. This folder can be simple copied to the board using SCP recursive command. NOTE : Some Pre-trained Applicatino | |||
= Support USB Camera = | |||
By default the Pretrained Applications are compiled to use the MIPI camera. The Pre-trained applications can be modified to use USB camera. This modification is only relevant to applications that support video. | |||
# Open the application folder src folder. For example [https://github.com/Ignitarium-Renesas/RZV2L_AiLibrary/tree/main/01_Head_count/Head_count_cam/src 01_Head_count/Head_count_cam/src] | |||
# Open the [https://github.com/Ignitarium-Renesas/RZV2L_AiLibrary/blob/main/01_Head_count/Head_count_cam/src/define.h define.h] header file. | |||
# Find the following line. Comment out the macro that defines INPUT_CORAL. | |||
<pre> | |||
/* Coral Camera support */ | |||
#define INPUT_CORAL | |||
</pre> | |||
= Sample Videos = | = Sample Videos = | ||
''NOTE: Memory Usage includes the Image Input, Inference Output, Inference Weights and Inference Parameters. Applications that use multiple AI Models are run sequentially'' | |||
{| style="width: 100%; border-style: none;" | {| style="width: 100%; border-style: none;" | ||
|- style="vertical-align: top;" | |- style="vertical-align: top;" | ||
|<br> <youtube width="480">YNcCCiSx9YM</youtube> <br> '''Head Count''' | | <br> | ||
| <br> <youtube width="480" >-fZypjgsBYo</youtube> <br> '''Line Count''' | === Head Count Application === | ||
|- | <br> Model : YoloV3 <br> Memory Usage: 235MB <br> Inference Input Shape : 416,416,3 <br> Inference Time: 348ms | ||
| <br> <youtube width="480" > | | <br> <youtube width="480">YNcCCiSx9YM</youtube> <br> '''Head Count''' | ||
| <br> <youtube width="480" >-DpAGb7q4pM</youtube> <br> '''Age and Gender Detection''' | |- style="vertical-align: top;" | ||
|- | | <br> | ||
| <br> <youtube width="480" >BOFdP1u-L7k</youtube> <br> '''Face Recognition''' | === Line Crossing Object Counting === | ||
<br> Model : TinyYoloV2 <br> Memory Usage: 52MB <br> Inference Input Shape : 416,416,3 <br> Inference Time: | |||
| <br> <youtube width="480">-fZypjgsBYo</youtube> <br> '''Line Count''' | |||
|- style="vertical-align: top;" | |||
| <br> | |||
=== Fall Detection === | |||
<br> Model : Tiny Yolov2 <br> Memory Usage: 52MB <br> Inference Input Shape : 256, 192, 3 <br> Inference Time: 59ms | |||
<br> Model : HRNET <br> Memory Usage: 129MB <br> Inference Input Shape : 256, 192, 3 <br> Inference Time: 163ms | |||
<br> Total Inference Time : 222ms | |||
| <br> <youtube width="480">4ALde_vP1lo</youtube> <br> '''Fall Detection''' | |||
|- style="vertical-align: top;" | |||
| <br> | |||
=== Age and Gender Detection === | |||
<br> Model : Custom age <br> Memory Usage: 24MB <br> Inference Input Shape : 416,416,3 <br> Inference Time: 10ms | |||
<br> Model : Custom gender <br> Memory Usage: 87MB <br> Inference Input Shape : 224,224,3 <br> Inference Time: 10ms | |||
<br> Model : Tiny YoloV2 <br> Memory Usage: 52MB <br> Inference Input Shape : 416,416,3 <br> Inference Time: 59ms | |||
<br> Total Inference Time: 79ms | |||
| <br> <youtube width="480">-DpAGb7q4pM</youtube> <br> '''Age and Gender Detection''' | |||
|- style="vertical-align: top;" | |||
| <br> | |||
=== Face Recognition, Spoofing, and Registration === | |||
<br> Model : Resnet50 <br> Memory Usage: 88MB <br> Inference Input Shape : 224,224,3 <br> Inference Time: 96ms | |||
| <br> <youtube width="480">BOFdP1u-L7k</youtube> <br> '''Face Recognition''' | |||
|- style="vertical-align: top;" | |||
| <br> | |||
=== Animal Detection === | |||
<br> Model : YoloV3 <br> Memory Usage: 236MB <br> Inference Input Shape : 416,416,3 <br> Inference Time: 360ms | |||
| <br> <youtube width="480">sJgDmCYcef4</youtube> <br> '''Animal Detection''' | |||
|- style="vertical-align: top;" | |||
| <br> | |||
=== Hand Gesture Recognition === | |||
<br> Model : Custom Pose Detector for Hand <br> Memory Usage: 91MB <br> Inference Input Shape : 256,256,3 <br> Inference Time: 256ms | |||
| <br> <youtube width="480">hP-Gr_Sq8a8</youtube> <br> '''Hand Gesture Recognition''' | |||
|- style="vertical-align: top;" | |||
| <br> | |||
=== Human Gaze Recognition === | |||
<br> Model : Resnet18 <br> Memory Usage: 38MB <br> Inference Input Shape : 416,416,3 <br> Inference Time: 33 | |||
<br> Model : Tiny YoloV2 <br> Memory Usage: 52MB <br> Inference Input Shape : 416,416,3 <br> Inference Time: 58 | |||
<br> Total Inference Time: 91ms | |||
| <br> <youtube width="480">X_eH5UcThrc</youtube> <br> '''Gaze Recognition''' | |||
|} | |} |
Latest revision as of 00:04, 20 April 2023
General Information
- Free, open-source based library of pre-trained AI applications available on github.
- ⭐Source Code: https://github.com/Ignitarium-Renesas/RZV2L_AiLibrary
- This Library has API functions for leveraging AI applications that will run on Renesas RZ/V2L Board. Currently this library has following sample applications:
- Human Head Counter
- Line crossing object Counter
- Elderly people fall detection (Work in progress)
- Safety helmet and vest detection
- Human age and gender detection (Work in progress)
- Face recognition and spoof detection (Work in progress
Addition Notes
The Pre-trained models include pre-compiled applications as well as AI Models translated to run on the DRP-AI hardware. These files are located in the "exe" folder for each Pre-trained Application. These precompiled application are compiled for the Renesas RZV2L EVK using the Coral MIPI Camera. This folder can be simple copied to the board using SCP recursive command. NOTE : Some Pre-trained Applicatino
Support USB Camera
By default the Pretrained Applications are compiled to use the MIPI camera. The Pre-trained applications can be modified to use USB camera. This modification is only relevant to applications that support video.
- Open the application folder src folder. For example 01_Head_count/Head_count_cam/src
- Open the define.h header file.
- Find the following line. Comment out the macro that defines INPUT_CORAL.
/* Coral Camera support */ #define INPUT_CORAL
Sample Videos
NOTE: Memory Usage includes the Image Input, Inference Output, Inference Weights and Inference Parameters. Applications that use multiple AI Models are run sequentially
Head Count Application
|
Head Count |
Line Crossing Object Counting
|
Line Count |
Fall Detection
|
Fall Detection |
Age and Gender Detection
|
Age and Gender Detection |
Face Recognition, Spoofing, and Registration
|
Face Recognition |
Animal Detection
|
Animal Detection |
Hand Gesture Recognition
|
Hand Gesture Recognition |
Human Gaze Recognition
|
Gaze Recognition |