DRP-AI Translator: Difference between revisions
(Fixed link) |
|||
(7 intermediate revisions by the same user not shown) | |||
Line 1: | Line 1: | ||
'''[https://www.renesas.com/us/en/products/microcontrollers-microprocessors/rz-arm-based-high-end-32-64-bit-mpus/drp-ai-translator Official DRP-AI Translator Website]''' | |||
= DRP-AI Reserved Memory = | = DRP-AI Reserved Memory = | ||
The DRP-AI Reserved memory is required for AI Inference processing. The memory areas is where the AI model binary, weights, bias, parameters, activations, and configuration files are located. | |||
=== DRP V5.00 to v7.00 === | |||
The latest version of the DRP library moved the reserved memory start address. Due to this change DRP-AI Applications that worked on DRP version 5.00 ( BSP v1.00 ) will result in errors with the latest DRP-AI drivers. | |||
To resolve the issue regenerate the DRP-AI models using the DRP-Translator. First modify the Address YAML configuration file to point to the new reserved memory. | |||
= DRP-AI Translator Address Map YAML = | = DRP-AI Translator Address Map YAML = | ||
The configuration file only requires the two modifications | The configuration file only requires the two modifications | ||
'''Step 1)''' Use the following template below<pre> | |||
<pre> | |||
name: "ALL" | name: "ALL" | ||
addr: 0x80000000 | addr: 0x80000000 | ||
lst_elemsp: | lst_elemsp: | ||
{ name: "data_in"} | - { name: "data_in"} | ||
{ name: "data"} | - { name: "data"} | ||
{ name: "data_out"} | - { name: "data_out"} | ||
{ name: "work"} | - { name: "work"} | ||
{ name: "weight"} | - { name: "weight"} | ||
{ name: "drp_config"} | - { name: "drp_config"} | ||
{ name: "drp_param"} | - { name: "drp_param"} | ||
{ name: "desc_aimac"} | - { name: "desc_aimac"} | ||
{ name: "desc_drp"} | - { name: "desc_drp"} | ||
</pre> | </pre> | ||
'''Step 2)''' Only the Address to the DRP-AI Reserved Memory needs to be set. For RZV2L this would be<pre> | |||
addr: 0x70000000 | |||
</pre> | |||
'''Step 3)''' After the DRP-AI Translator is run a new address yaml and text files is generated. Below is an example generated ymal for TinyYoloV2.. | |||
<pre> | <pre> | ||
addr: | name: ALL | ||
addr: 0x7000_0000 | |||
size: 0x30f_68c0 | |||
lst_elemsp: | |||
- { name: data_in, addr: 0x7000_0000, size: 0x9_6000 } | |||
- { name: data, addr: 0x7009_6000, size: 0x103_9c16 } | |||
- { name: data_out, addr: 0x710c_fc40, size: 0x1_4a14 } | |||
- { name: work, addr: 0x710e_4680, size: 0xd000 } | |||
- { name: weight, addr: 0x710f_1680, size: 0x1e4_280c } | |||
- { name: drp_config, addr: 0x72f3_3ec0, size: 0x1a_6280 } | |||
- { name: drp_param, addr: 0x730d_a140, size: 0x270 } | |||
- { name: desc_aimac, addr: 0x730d_a3c0, size: 0x1_c1e0 } | |||
- { name: desc_drp, addr: 0x730f_65c0, size: 0x300 } | |||
</pre> | </pre> | ||
= DRP-AI Translator Pre/Post Processing YAML = | =DRP-AI Translator Pre/Post Processing YAML= | ||
The section we explain the required fields that are needed for pre/post yaml file. | |||
For detailed information on the YAML files please refer to the DR--AI Translator Manual. | |||
=DRP-AI Implementation Guide= | |||
Renesas Provides Implementation Guides for Renesas Implemented AI Models can be found here. These guides show how to convert pyTorch AI model to ONNX Format, Use the DRP-AI Translator, and finally Execute the DRP-AI translated mode on the target board. | |||
[[File:Figure 1.png|thumb|Implementation Guide Overview]] | |||
= DRP-AI Translation Guide = | |||
This section describe the translation procedure for the following AI algorithms that are not fully supported on the DRP-AI Translator. For the list of supported AI operation please refer to the DRP-AI Translator User Manual section 4.1.1 ONNX Operators. | |||
* [[Translation procedure for YOLOv5|Translation procedure for '''YOLOv5''']] | |||
* [[Translation procedure for YOLOX|Translation procedure for '''YOLOX''']] | |||
= DRP-AI Translator Version Support = | |||
[[File:DRP-AI Translator Version Support.png|left|thumb|800x800px|AI Translator Versions]] |
Latest revision as of 21:02, 31 May 2023
Official DRP-AI Translator Website
DRP-AI Reserved Memory
The DRP-AI Reserved memory is required for AI Inference processing. The memory areas is where the AI model binary, weights, bias, parameters, activations, and configuration files are located.
DRP V5.00 to v7.00
The latest version of the DRP library moved the reserved memory start address. Due to this change DRP-AI Applications that worked on DRP version 5.00 ( BSP v1.00 ) will result in errors with the latest DRP-AI drivers.
To resolve the issue regenerate the DRP-AI models using the DRP-Translator. First modify the Address YAML configuration file to point to the new reserved memory.
DRP-AI Translator Address Map YAML
The configuration file only requires the two modifications
Step 1) Use the following template below
name: "ALL" addr: 0x80000000 lst_elemsp: - { name: "data_in"} - { name: "data"} - { name: "data_out"} - { name: "work"} - { name: "weight"} - { name: "drp_config"} - { name: "drp_param"} - { name: "desc_aimac"} - { name: "desc_drp"}
Step 2) Only the Address to the DRP-AI Reserved Memory needs to be set. For RZV2L this would be
addr: 0x70000000
Step 3) After the DRP-AI Translator is run a new address yaml and text files is generated. Below is an example generated ymal for TinyYoloV2..
name: ALL addr: 0x7000_0000 size: 0x30f_68c0 lst_elemsp: - { name: data_in, addr: 0x7000_0000, size: 0x9_6000 } - { name: data, addr: 0x7009_6000, size: 0x103_9c16 } - { name: data_out, addr: 0x710c_fc40, size: 0x1_4a14 } - { name: work, addr: 0x710e_4680, size: 0xd000 } - { name: weight, addr: 0x710f_1680, size: 0x1e4_280c } - { name: drp_config, addr: 0x72f3_3ec0, size: 0x1a_6280 } - { name: drp_param, addr: 0x730d_a140, size: 0x270 } - { name: desc_aimac, addr: 0x730d_a3c0, size: 0x1_c1e0 } - { name: desc_drp, addr: 0x730f_65c0, size: 0x300 }
DRP-AI Translator Pre/Post Processing YAML
The section we explain the required fields that are needed for pre/post yaml file.
For detailed information on the YAML files please refer to the DR--AI Translator Manual.
DRP-AI Implementation Guide
Renesas Provides Implementation Guides for Renesas Implemented AI Models can be found here. These guides show how to convert pyTorch AI model to ONNX Format, Use the DRP-AI Translator, and finally Execute the DRP-AI translated mode on the target board.
DRP-AI Translation Guide
This section describe the translation procedure for the following AI algorithms that are not fully supported on the DRP-AI Translator. For the list of supported AI operation please refer to the DRP-AI Translator User Manual section 4.1.1 ONNX Operators.