Deploying Kendryte Training-Platform Models on RT-Smart#

Attention

This deployment flow uses the same single-camera dual-channel pattern as the task templates. Refer to single_model_example.md for single-model tasks and double_model_example.md for OCR detection plus recognition.

Overview#

cloudplat_deploy_code packages the deployment code for models exported from the Kendryte online training platform. Build the executable under rtos_sdk, then deploy the exported model package to the board.

For the firmware build flow, see:

how_to_build

Source Description#

cloudplat_deploy_code implements all task types currently supported by the training platform:

image classification
object detection
semantic segmentation
OCR detection
OCR recognition
OCR detection + recognition
metric learning
multi-label classification

The common logic for model inference, preprocessing helpers, config parsing, and result drawing is placed in common_files. Each task directory then implements the task-specific inference flow.

Code Structure#

The directory layout is:

cloudplat_deploy_code_linux
├── common_files
├── classification
├── detection
├── segmentation
├── ocr_detection
├── ocr_recognition
├── ocr
├── metric_learning
├── multilabel_classification
├── utils
│   └── SourceHanSansSC-Normal-Min.ttf
├── libs
├── CMakeLists.txt
└── build.sh

Each task directory contains the application entry, task-specific preprocess and postprocess logic, and deployment helpers for that task type.

Build Notes#

Parameter Configuration#

You can configure the display-related parameters in:

common_files/setting.h

The main macros are:

Macro	Description
`ISP_WIDTH`	ISP output width
`ISP_HEIGHT`	ISP output height
`DISPLAY_MODE`	Display mode, `0`: `1920 x 1080` LT9611, `1`: `800 x 480` ST7701
`DISPLAY_WIDTH`	Display width
`DISPLAY_HEIGHT`	Display height
`AI_FRAME_WIDTH`	AI inference frame width
`AI_FRAME_HEIGHT`	AI inference frame height
`AI_FRAME_CHANNEL`	AI inference frame channel count
`USE_OSD`	Whether to enable OSD, `0`: disable, `1`: enable
`OSD_WIDTH`	OSD layer width for AI results
`OSD_HEIGHT`	OSD layer height for AI results
`OSD_CHANNEL`	OSD layer channel count

These settings are mainly used to configure the board-side display path. As in the other AI guides, the AI frame resolution is the camera-side frame before preprocessing, while the model input resolution is the tensor size after preprocessing.

Source Build#

Enter src/rtsmart/examples/ai/cloudplat_deploy_code and run:

cd cloudplat_deploy_code

./build_app.sh

./build_app.sh classification
./build_app.sh detection

If you want to build all task deployment files directly, you can also run:

make -j

The build artifacts are collected in k230_bin.

Board Deployment#

Copy the generated .elf files, the bundled font file, the exported kmodel, deploy_config.json, and any test images to one directory on the board. Then run the command for the corresponding task.

# Classification, video inference. Input q then Enter to exit.
./classification.elf deploy_config.json None 0

# Classification, image inference.
./classification.elf deploy_config.json test.jpg 0

# Detection, video inference.
./detection.elf deploy_config.json None 0

# Detection, image inference.
./detection.elf deploy_config.json test.jpg 0

# Semantic segmentation, video inference.
./segmentation.elf deploy_config.json None 0

# Semantic segmentation, image inference.
./segmentation.elf deploy_config.json test.jpg 0

# OCR detection, video inference.
./ocr_detection.elf deploy_config.json None 0

# OCR detection, image inference.
./ocr_detection.elf deploy_config.json test.jpg 0

# OCR recognition, image inference only.
./ocr_recognition.elf deploy_config.json test.jpg 0

# OCR detection + recognition, video inference.
./ocr.elf ocrdet_deploy_config.json ocrrec_deploy_config.json None 0

# OCR detection + recognition, image inference.
./ocr.elf ocrdet_deploy_config.json ocrrec_deploy_config.json test.jpg 0

# Metric learning, video inference.
./metric_learning.elf deploy_config.json None 0

# Metric learning, image inference.
./metric_learning.elf deploy_config.json test.jpg 0

# Multi-label classification, video inference.
./multilabel_classification.elf deploy_config.json None 0

# Multi-label classification, image inference.
./multilabel_classification.elf deploy_config.json test.jpg 0

The last argument is the debug level and typically follows the same convention as the other AI samples:

0: no debug output
1: timing-oriented debug output
2: verbose debug output