RZ-V/RZ-V2H EVK: Difference between revisions
(→Prebuilt Images: Added information about the contents of the AI SDK) |
(→DRP-AI Tools: Added more information about the DRP Tools) |
||
| Line 63: | Line 63: | ||
= DRP-AI Tools = | = DRP-AI Tools = | ||
[[File:DRP Translation Flow.png|alt=Show the basic flow of AI Model Translation to the DRP AI hardware.|center|frame|Show the basic flow of AI Model Translation to the DRP AI hardware.]] | |||
⚠️ The DRP-AI Tools will only work in Ubuntu 20.04 | ⚠️ The DRP-AI Tools will only work in Ubuntu 20.04 | ||
== RZ/V2H DRP-AI Translator and Quantization Tool == | == RZ/V2H DRP-AI TVM == | ||
'''[https://github.com/renesas-rz/rzv_drp-ai_tvm TVM Repository]''' | |||
'''[[DRP-AI TVM|DRP-AI TVM wik]]<nowiki/>i''' | |||
The Renesas TVM is the DRP-AI translator tool that converts standard ML Frameworks ( Pytorch, Tensorflow, ONNX, .. ) to DRP-AI Hardware object files. The TVM for RZ/V2H includes the following tools. This is the recommended tool for '''AI Model Translation to the DRP-AI'''. | |||
* Inference Pre-processing Configuration | |||
* Quantization Tool | |||
* Translation Compiler | |||
=== RZ/V2H DRP-AI Translator and Quantization Tool === | |||
'''[https://www.renesas.com/us/en/software-tool/drp-ai-translator-i8 Official Site]''' | '''[https://www.renesas.com/us/en/software-tool/drp-ai-translator-i8 Official Site]''' | ||
This tool quantize and translates AI Models exported in the ONNX format to the RZ/V2H DRP-AI (INT8) hardware. This tool includes the following | This tool iused by the '''TVM''' for translation of AI Models. The tool supports quantize and translates AI Models exported in the ONNX format to the RZ/V2H DRP-AI (INT8) hardware. This tool includes the following | ||
* '''RZ/V2H DRP-AI Translator''' | * '''RZ/V2H DRP-AI Translator''' | ||
| Line 84: | Line 96: | ||
* Customer Dataset ( does not need to be the full dataset ) | * Customer Dataset ( does not need to be the full dataset ) | ||
* Function to parse the mode | * Function to parse the mode | ||
== RZ/V2H Expansion Pack (Pruning Tool ) == | == RZ/V2H Expansion Pack (Pruning Tool ) == | ||
| Line 94: | Line 101: | ||
'''[https://www.renesas.com/us/en/software-tool/drp-ai-extension-pack-pruning-tool Official Site]''' | '''[https://www.renesas.com/us/en/software-tool/drp-ai-extension-pack-pruning-tool Official Site]''' | ||
= Getting Started = | = Getting Started = | ||
Revision as of 01:40, 29 March 2024
General Information
- Official RZ/V2H Device Website *
- Product Page
- Please review the **Documentation & Downloads** section
- Evaluation Board Documentation
- Software Downloads
- Check "Software & Tool Pages" section on on this page
- RZ/V2H Linux Package
- Mali Graphic Library
- Video Codec Library
- ARM Mali Full ISP
- RZ/V2L DRP-AI Support Package
- RZ/V2H Cortex-M33 and Cortex-CR8 Multi-OS Package (optional)
- Robotic Operating System (ROS)
- RZ/V2H Cortex CM33 FSP
Prebuilt Images
To help expedite evaluation of the RZ/V2H we provide Pre-build Images of the RZ/V2H VLP.
AI SDK Contents
eSD Card Image
The eSD Card Image is a prebuilt yocto image that contains the following yocto packages
- VLP Linux
- ARM Mali Graphics ( limited license )
- Codec ( limited license )
- DRP-AI Driver
- OpenCVA ( DRP Accelerated )
The eSD Image DOES NOT include the following
- e-CON MIPI Camera ( Contact e-CON Systems )
- AI SDK Demos ( Comming Soon )
TVM Translator Environment
The TVM Translator is the recommended translator to translate AI models for the DRP-AI Driver. All the AI SDK demos use the TVM Translator. Included in the AI SDK is the DRP-AI Translator. This translator is needed for TVM installation and does not need to be installed. Installation using Docker is recommended because the TVM Environment requires a PC running Ubuntu 20.04. For information about Docker setup see here.
For information about using the TVM Translator see the Official Site here.
NOTE: Once the AI SDK Environment is setup the TVM will be installed in the the root dirctory /drp_tvm_sdk
Available Demos
DRP-AI
- vSLAM (Contact Sales Representative for demo)
- DRP-AI Demos
- Included with the DRP-AI Package. AI models include
- Classification (Resnet50)
- Object Detection ( YoloX and Yolov2 )
- Segmentation ( DeepLabv2 )
- Included with the DRP-AI Package. AI models include
- Renesas RZ/V2H AI SDK
DRP-AI Tools
⚠️ The DRP-AI Tools will only work in Ubuntu 20.04
RZ/V2H DRP-AI TVM
The Renesas TVM is the DRP-AI translator tool that converts standard ML Frameworks ( Pytorch, Tensorflow, ONNX, .. ) to DRP-AI Hardware object files. The TVM for RZ/V2H includes the following tools. This is the recommended tool for AI Model Translation to the DRP-AI.
- Inference Pre-processing Configuration
- Quantization Tool
- Translation Compiler
RZ/V2H DRP-AI Translator and Quantization Tool
This tool iused by the TVM for translation of AI Models. The tool supports quantize and translates AI Models exported in the ONNX format to the RZ/V2H DRP-AI (INT8) hardware. This tool includes the following
- RZ/V2H DRP-AI Translator
- DRP-AI Quantization Tool
The RZV2H DRP-AI MAC only performs INT8 operations, so AI Models must be quantized before translation.
AI Models must use only the supported AI operations listed in the RZ\V2H DRP-AI Translator Manual ( section 4.1). For AI models that have un-supported operations please use the Renesas DRP-AI TVM Tool.
Retraining Quantized Models
It is recommended to retrains a model after quantization to reduce changes to the inference accuracy . For that purpose the Quantization tool includes optional parameters that point to the data set and dataset parsing function.
Beside the AI Model below are additional items needed to properly quantize a model.
- Customer Dataset ( does not need to be the full dataset )
- Function to parse the mode
RZ/V2H Expansion Pack (Pruning Tool )
The DRP-AI MAC supports sparse arrays (unstructured pruning). The expansion pack tool are python libraries that can be imported into the training environment.
Getting Started
Hardware
Power Supply
Recommended Power Supplies ( Recommend 100Ws )
| Description | Link | Comments | |
|---|---|---|---|
| 🙂 | Amazon Basics | Link | 2xUSB-C, 2xUSB,
1x65-100W, 1x18W, 2x12W |
| 🙂 | Nekteck | Link | 1x USBC 100W |
+LCD Monitors
| Resolution | Touchscreen | Description | Link | Comments | |
|---|---|---|---|---|---|
| 🙂 | 1920x1080 | Yes | Wimaxit 14 Inch | 🛒 buy | This is a very nice screen. Good color and easy to press the icons. The LCD input is mini-HDMI and comes with a mini-HDMI to HDMI cable. The RZ board is micro-HDMI. You will also need to buy a HDMI to micro-HDMI adapter. Or, buy a mini-HDMI to HDMI-micro cable. Note the the micro-USB connection on the LCD for touchscreen does not work, but the USB-C connector does work for touchscreen. You will simply need a USB-C to USB-A cable or adapter. |
| 🙂 | 1920x1080 | Yes | ELECROW 10.1 Inch | 🛒 buy | This is a very nice screen. You will also need to buy a HDMI to micro-HDMI cable. |
| 🙂 | 1280x800 | Yes | SunFounder 10.1 inch | 🛒 buy | All the cables that you will need are included. |
| 🙂 | 1024x600 | Yes | Ingcool 7 inch 7IP-CAPLCD |
🛒 buy | If you attach the 'Touch' USB connection on the LCD to the RZ/G2L board, it will both power the LCD board as well as enable touch support. This kit comes with HDMI cables. |
| 🙂 | 800x480 | Yes | GeeekPi 5 Inch EP-0081 |
🛒 buy | HDMI cable not included. |
MIPI Camera
Supported MIPI Cameras
| Resolutions | Description | Link | Comments | |
|---|---|---|---|---|
| 🙂 | VGA ( 640x480) , HD ( 1270x720), FHD(1920,1080)
Format YUYV FPS 30 and 60 |
e-CAM22_CURZH | Link | This Camera was designed for the EVK MIPI Connector Interface. Uses the Sony Starvis I IMX462. |