xing/amlnn-model-playground
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

feat: Update NNSDK path and library linking in CMake configurations and Android build script, and add a new script to build all Android examples.

Browse source
This commit is contained in:
dian.yuan 2026-02-24 19:41:14 +08:00
parent 79a2bd27f6
commit 21105e3db7
35 changed files with 1438 additions and 1222 deletions
Download patch file Download diff file Expand all files Collapse all files

47
examples/build-android-all.sh
View file

@ -41,20 +41,43 @@ while getopts 'a:h' opt; do
esac
done
if [ -z "${ANDROID_NDK_PATH}" ]; then
if [ -n "${ANDROID_NDK}" ]; then
ANDROID_NDK_PATH=${ANDROID_NDK}
elif [ -n "${ANDROID_NDK_HOME}" ]; then
ANDROID_NDK_PATH=${ANDROID_NDK_HOME}
else
echo "Error: ANDROID_NDK_PATH is not set."
echo "Please set ANDROID_NDK_PATH to your Android NDK directory."
SCRIPT_DIR=$(cd "$(dirname $0)" && pwd)
# Priority 1: Environment variable (recommended)
if [ -n "$AMLNN_HOME" ]; then
if [ ! -d "$AMLNN_HOME/nn_runtime" ]; then
echo "Error: AMLNN_HOME is set to '$AMLNN_HOME' but nn_runtime was not found there."
echo "Please check your AMLNN_HOME path."
exit 1
fi
RUNTIME_PATH="$AMLNN_HOME/nn_runtime"
echo "Priority 1: Using AMLNN_HOME from environment: $AMLNN_HOME"
# Priority 3: Fallback to sibling amlnn-toolkit (compatibility)
elif [ -d "${SCRIPT_DIR}/../../amlnn-toolkit/nn_runtime" ]; then
export AMLNN_HOME="$(cd "${SCRIPT_DIR}/../../amlnn-toolkit" && pwd)"
RUNTIME_PATH="$AMLNN_HOME/nn_runtime"
echo "Priority 3: Using sibling amlnn-toolkit as fallback: $AMLNN_HOME"
elif [ -d "${SCRIPT_DIR}/../../amlnn-toolkit-a/nn_runtime" ]; then
export AMLNN_HOME="$(cd "${SCRIPT_DIR}/../../amlnn-toolkit-a" && pwd)"
RUNTIME_PATH="$AMLNN_HOME/nn_runtime"
echo "Priority 3: Using sibling amlnn-toolkit-a as fallback: $AMLNN_HOME"
else
echo ""
echo "Error: AMLNN SDK not found."
echo ""
echo "Please do one of the following:"
echo ""
echo " Option A (recommended) – set AMLNN_HOME:"
echo " export AMLNN_HOME=/path/to/amlnn-toolkit"
echo " ./build-android-all.sh"
echo ""
echo " Option B – clone amlnn-toolkit as a sibling directory:"
echo " git clone git@github.com:Amlogic-NN/amlnn-toolkit.git ../../../amlnn-toolkit"
echo " ./build-android-all.sh"
echo ""
exit 1
fi
SCRIPT_DIR=$(cd "$(dirname $0)" && pwd)
echo "============================================"
echo "Building all Android examples"
echo "NDK_PATH: ${ANDROID_NDK_PATH}"
@ -93,6 +116,10 @@ for EXAMPLE in "${EXAMPLES[@]}"; do
echo "Building: ${EXAMPLE}"
echo "--------------------------------------------"
# Clean previous build to avoid stale CMake cache
echo "Cleaning: ${EXAMPLE_DIR}/build/android"
rm -rf "${EXAMPLE_DIR}/build/android"
if bash "${BUILD_SCRIPT}" -a "${TARGET_ABI}"; then
SUCCEEDED+=("${EXAMPLE}")
echo "[SUCCESS] ${EXAMPLE}"

14
examples/clip/README.md
View file

@ -18,6 +18,20 @@ TO DO
#### 1. Compile
#### AMLNN SDK Setup
Resolve the AMLNN nnsdk dependency using one of the following methods:
- **Priority 1 – Environment variable (recommended)**
```bash
export AMLNN_HOME=/path/to/amlnn-toolkit
```
- **Priority 3 – Sibling directory fallback** *(automatic)*
Place `amlnn-toolkit` as a sibling to `amlnn-model-playground`:
```bash
git clone git@github.com:Amlogic-NN/amlnn-toolkit.git ../amlnn-toolkit
```
**Prerequisites:**
- Android NDK (r25e recommended)
- `ANDROID_NDK_PATH` environment variable set

3
examples/clip/cpp/build-android.sh
View file

@ -65,7 +65,8 @@ echo "BUILD_DIR: ${BUILD_DIR}"
mkdir -p ${BUILD_DIR}
cd ${BUILD_DIR}
cmake ../../src \
cmake -Wno-dev ../../src \
-DAMLNN_HOME=${AMLNN_HOME:-} \
-DCMAKE_TOOLCHAIN_FILE=${ANDROID_NDK_PATH}/build/cmake/android.toolchain.cmake \
-DANDROID_ABI=${TARGET_ABI} \
-DANDROID_PLATFORM=android-24 \

22
examples/clip/cpp/src/CMakeLists.txt
View file

@ -1,16 +1,13 @@
cmake_minimum_required(VERSION 3.5)
cmake_minimum_required(VERSION 3.10...3.27)
project(clip_demo)
set(CMAKE_CXX_STANDARD 17)
# Set NNSDK path
if(NOT DEFINED NNSDK_DIR)
set(NNSDK_DIR "${CMAKE_SOURCE_DIR}/../../../../../amlnn-toolkit/nn_runtime/nnsdk")
endif()
set(NNSDK_ROOT "${NNSDK_DIR}")
message(STATUS "NNSDK_ROOT: ${NNSDK_ROOT}")
list(APPEND CMAKE_MODULE_PATH "${CMAKE_SOURCE_DIR}/../../../../cmake")
find_package(AMLNN REQUIRED)
include_directories(${AMLNN_INCLUDE_DIR})
link_directories(${AMLNN_LIBRARY_DIR})
include_directories(${NNSDK_ROOT}/include)
include_directories(${CMAKE_SOURCE_DIR}/../../../../common)
# Set 3rdparty path
@ -22,15 +19,8 @@ include_directories(${3RDPARTY_DIR}/stb_image)
include_directories(${3RDPARTY_DIR}/json)
if(CMAKE_SYSTEM_NAME STREQUAL "Android")
if (ANDROID_ABI STREQUAL "arm64-v8a")
link_directories(${NNSDK_ROOT}/android/arm64-v8a)
else()
link_directories(${NNSDK_ROOT}/android/armeabi-v7a)
endif()
# Android needs log
link_libraries(log)
elseif(CMAKE_SYSTEM_NAME STREQUAL "Linux")
link_directories(${NNSDK_ROOT}/linux/yocto/aarch64-poky-linux)
endif()
add_executable(${PROJECT_NAME}
@ -41,7 +31,7 @@ add_executable(${PROJECT_NAME}
)
target_link_libraries(${PROJECT_NAME}
nnsdk
${AMLNN_LIBRARY}
dl
m
)

250
examples/mobilenet/README.md
View file

@ -1,118 +1,132 @@
## Model Description
This model is converted from MobileNetV2 pretrained weights
originally released by Google under the Apache License 2.0.
Original model:
- Architecture: MobileNetV2
- Source: TensorFlow / Keras official implementation
The model has been converted and optimized into ADLA format
for deployment on Amlogic NPU platforms.
## Demo Run
### CPP
#### 1. Compile
**Prerequisites:**
- Android NDK (r25e recommended)
- `ANDROID_NDK_PATH` environment variable set
**Build:**
```bash
# Build for arm64-v8a
cd examples/mobilenet/cpp
./build-android.sh -a arm64-v8a
```
The executable will be generated at `build/android/mobilenet_v2_demo` (Note: executable name may vary, verify in build folder).
#### 2. Run
```bash
# Push executable to device
adb push build/android/mobilenet_v2_demo /data/local/tmp/
adb push model/mobilenet_v2_1.0_224_quant_A311D2.adla /data/local/tmp/
adb push model/cat_224x224.jpg /data/local/tmp/
adb push model/labels.txt /data/local/tmp/
# Run on device
adb shell
cd /data/local/tmp
chmod +x mobilenet_v2_demo
export LD_LIBRARY_PATH=/vendor/lib64 or (/vendor/lib)
# Usage: ./mobilenet_v2_demo <model_path> <image_path> <labels_path>
./mobilenet_v2_demo mobilenet_v2_1.0_224_quant_A311D2.adla cat_224x224.jpg labels.txt
```
**Note:** Replace `mobilenet_v2_1.0_224_quant_A311D2.adla` with your actual model file path.
### Python
**Prerequisites:**
- Python 3.10
- Required packages: `numpy`, `Pillow`, `amlnnlite`
**Install dependencies:**
```bash
pip install numpy Pillow amlnnlite-1.0.0-cp310-cp310-linux_aarch64.whl
```
**Run on device:**
```bash
# Basic usage
python mobilenetv2.py --model-path ./mobilenet_v2_1.0_224_quant_A311D2.adla
# Run with performance testing (100 cycles)
python mobilenetv2.py --model-path ./mobilenet_v2_1.0_224_quant_A311D2.adla --run-cycles 100
```
The script will automatically process all image files (`.jpg`, `.jpeg`, `.png`, `.bmp`) in the current directory and display top-5 classification results for each image.
## Results
The program will print the top-5 classification results with probabilities for each processed image.
**Example output:**
```
# python demo result
============================================================
Processing image 1/3: dog_224x224.jpg
============================================================
Top-5 Classification Results:
1. Shih-Tzu (probability: 0.9239)
2. Pekinese (probability: 0.0476)
3. Lhasa (probability: 0.0263)
4. Brabancon griffon (probability: 0.0004)
5. Dandie Dinmont (probability: 0.0003)
============================================================
Processing image 2/3: cat_224x224.jpg
============================================================
Top-5 Classification Results:
1. tiger cat (probability: 0.4774)
2. tabby (probability: 0.4324)
3. Egyptian cat (probability: 0.0542)
4. lynx (probability: 0.0150)
5. Persian cat (probability: 0.0025)
============================================================
Processing image 3/3: fish_224x224.jpeg
============================================================
Top-5 Classification Results:
1. goldfish (probability: 0.9998)
2. conch (probability: 0.0001)
3. trifle (probability: 0.0000)
4. axolotl (probability: 0.0000)
5. American lobster (probability: 0.0000)
```
The classification results show the model's confidence scores (probabilities) for each detected class, with the highest probability indicating the most likely classification.
## Model Description
This model is converted from MobileNetV2 pretrained weights
originally released by Google under the Apache License 2.0.
Original model:
- Architecture: MobileNetV2
- Source: TensorFlow / Keras official implementation
The model has been converted and optimized into ADLA format
for deployment on Amlogic NPU platforms.
## Demo Run
### CPP
#### 1. Compile
#### AMLNN SDK Setup
Resolve the AMLNN nnsdk dependency using one of the following methods:
- **Priority 1 – Environment variable (recommended)**
```bash
export AMLNN_HOME=/path/to/amlnn-toolkit
```
- **Priority 3 – Sibling directory fallback** *(automatic)*
Place `amlnn-toolkit` as a sibling to `amlnn-model-playground`:
```bash
git clone git@github.com:Amlogic-NN/amlnn-toolkit.git ../amlnn-toolkit
```
**Prerequisites:**
- Android NDK (r25e recommended)
- `ANDROID_NDK_PATH` environment variable set
**Build:**
```bash
# Build for arm64-v8a
cd examples/mobilenet/cpp
./build-android.sh -a arm64-v8a
```
The executable will be generated at `build/android/mobilenet_v2_demo` (Note: executable name may vary, verify in build folder).
#### 2. Run
```bash
# Push executable to device
adb push build/android/mobilenet_v2_demo /data/local/tmp/
adb push model/mobilenet_v2_1.0_224_quant_A311D2.adla /data/local/tmp/
adb push model/cat_224x224.jpg /data/local/tmp/
adb push model/labels.txt /data/local/tmp/
# Run on device
adb shell
cd /data/local/tmp
chmod +x mobilenet_v2_demo
export LD_LIBRARY_PATH=/vendor/lib64 or (/vendor/lib)
# Usage: ./mobilenet_v2_demo <model_path> <image_path> <labels_path>
./mobilenet_v2_demo mobilenet_v2_1.0_224_quant_A311D2.adla cat_224x224.jpg labels.txt
```
**Note:** Replace `mobilenet_v2_1.0_224_quant_A311D2.adla` with your actual model file path.
### Python
**Prerequisites:**
- Python 3.10
- Required packages: `numpy`, `Pillow`, `amlnnlite`
**Install dependencies:**
```bash
pip install numpy Pillow amlnnlite-1.0.0-cp310-cp310-linux_aarch64.whl
```
**Run on device:**
```bash
# Basic usage
python mobilenetv2.py --model-path ./mobilenet_v2_1.0_224_quant_A311D2.adla
# Run with performance testing (100 cycles)
python mobilenetv2.py --model-path ./mobilenet_v2_1.0_224_quant_A311D2.adla --run-cycles 100
```
The script will automatically process all image files (`.jpg`, `.jpeg`, `.png`, `.bmp`) in the current directory and display top-5 classification results for each image.
## Results
The program will print the top-5 classification results with probabilities for each processed image.
**Example output:**
```
# python demo result
============================================================
Processing image 1/3: dog_224x224.jpg
============================================================
Top-5 Classification Results:
1. Shih-Tzu (probability: 0.9239)
2. Pekinese (probability: 0.0476)
3. Lhasa (probability: 0.0263)
4. Brabancon griffon (probability: 0.0004)
5. Dandie Dinmont (probability: 0.0003)
============================================================
Processing image 2/3: cat_224x224.jpg
============================================================
Top-5 Classification Results:
1. tiger cat (probability: 0.4774)
2. tabby (probability: 0.4324)
3. Egyptian cat (probability: 0.0542)
4. lynx (probability: 0.0150)
5. Persian cat (probability: 0.0025)
============================================================
Processing image 3/3: fish_224x224.jpeg
============================================================
Top-5 Classification Results:
1. goldfish (probability: 0.9998)
2. conch (probability: 0.0001)
3. trifle (probability: 0.0000)
4. axolotl (probability: 0.0000)
5. American lobster (probability: 0.0000)
```
The classification results show the model's confidence scores (probabilities) for each detected class, with the highest probability indicating the most likely classification.

3
examples/mobilenet/cpp/build-android.sh
View file

@ -49,7 +49,8 @@ echo "BUILD_DIR: ${BUILD_DIR}"
mkdir -p ${BUILD_DIR}
cd ${BUILD_DIR}
cmake ../../src \
cmake -Wno-dev ../../src \
-DAMLNN_HOME=${AMLNN_HOME:-} \
-DCMAKE_TOOLCHAIN_FILE=${ANDROID_NDK_PATH}/build/cmake/android.toolchain.cmake \
-DANDROID_ABI=${TARGET_ABI} \
-DANDROID_PLATFORM=android-24 \

22
examples/mobilenet/cpp/src/CMakeLists.txt
View file

@ -1,31 +1,21 @@
cmake_minimum_required(VERSION 3.5)
cmake_minimum_required(VERSION 3.10...3.27)
project(mobilenet_v2_demo)
set(CMAKE_CXX_STANDARD 17)
# Set NNSDK path
if(NOT DEFINED NNSDK_DIR)
set(NNSDK_DIR "${CMAKE_SOURCE_DIR}/../../../../../amlnn-toolkit/nn_runtime/nnsdk")
endif()
set(NNSDK_ROOT "${NNSDK_DIR}")
message(STATUS "NNSDK_ROOT: ${NNSDK_ROOT}")
list(APPEND CMAKE_MODULE_PATH "${CMAKE_SOURCE_DIR}/../../../../cmake")
find_package(AMLNN REQUIRED)
include_directories(${AMLNN_INCLUDE_DIR})
link_directories(${AMLNN_LIBRARY_DIR})
include_directories(${NNSDK_ROOT}/include)
include_directories(${CMAKE_SOURCE_DIR}/../../../../common)
# Set dependency path
set(3RDPARTY_DIR "${CMAKE_SOURCE_DIR}/../../../../dependency")
if(CMAKE_SYSTEM_NAME STREQUAL "Android")
if (ANDROID_ABI STREQUAL "arm64-v8a")
link_directories(${NNSDK_ROOT}/android/arm64-v8a)
else()
link_directories(${NNSDK_ROOT}/android/armeabi-v7a)
endif()
# Android needs log
link_libraries(log)
elseif(CMAKE_SYSTEM_NAME STREQUAL "Linux")
link_directories(${NNSDK_ROOT}/linux/yocto/aarch64-poky-linux)
endif()
# Find OpenCV
@ -40,5 +30,5 @@ add_executable(mobilenet_v2_demo
target_link_libraries(mobilenet_v2_demo
${OpenCV_LIBS}
nnsdk
${AMLNN_LIBRARY}
)

34
examples/ppocr-det/README.md
View file

@ -0,0 +1,34 @@
# PaddleOCR Detection
## 4. Demo Run
### CPP
#### 1. Compile
#### AMLNN SDK Setup
Resolve the AMLNN nnsdk dependency using one of the following methods:
- **Priority 1 – Environment variable (recommended)**
```bash
export AMLNN_HOME=/path/to/amlnn-toolkit
```
- **Priority 3 – Sibling directory fallback** *(automatic)*
Place `amlnn-toolkit` as a sibling to `amlnn-model-playground`:
```bash
git clone git@github.com:Amlogic-NN/amlnn-toolkit.git ../amlnn-toolkit
```
**Prerequisites:**
- Android NDK r25c
- `ANDROID_NDK_PATH` environment variable set
**Build:**
```bash
# Build for arm64-v8a
cd examples/ppocr-det/cpp
./build-android.sh -a arm64-v8a
```
The executable will be generated in `build/android/`.

3
examples/ppocr-det/cpp/build-android.sh
View file

@ -65,7 +65,8 @@ echo "BUILD_DIR: ${BUILD_DIR}"
mkdir -p ${BUILD_DIR}
cd ${BUILD_DIR}
cmake ../../src \
cmake -Wno-dev ../../src \
-DAMLNN_HOME=${AMLNN_HOME:-} \
-DCMAKE_TOOLCHAIN_FILE=${ANDROID_NDK_PATH}/build/cmake/android.toolchain.cmake \
-DANDROID_ABI=${TARGET_ABI} \
-DANDROID_PLATFORM=android-24 \

86
examples/ppocr-det/cpp/src/CMakeLists.txt
View file

@ -1,48 +1,38 @@
cmake_minimum_required(VERSION 3.5)
project(yolo_world_demo)
set(CMAKE_CXX_STANDARD 17)
# Set NNSDK path
if(NOT DEFINED NNSDK_DIR)
set(NNSDK_DIR "${CMAKE_SOURCE_DIR}/../../../../../amlnn-toolkit/nn_runtime/nnsdk")
endif()
set(NNSDK_ROOT "${NNSDK_DIR}")
message(STATUS "NNSDK_ROOT: ${NNSDK_ROOT}")
include_directories(${NNSDK_ROOT}/include)
include_directories(${CMAKE_SOURCE_DIR}/../../../../common)
# Set 3rdparty path
set(3RDPARTY_DIR "${CMAKE_SOURCE_DIR}/../../../../dependency")
if(CMAKE_SYSTEM_NAME STREQUAL "Android")
if (ANDROID_ABI STREQUAL "arm64-v8a")
link_directories(${NNSDK_ROOT}/android/arm64-v8a)
else()
link_directories(${NNSDK_ROOT}/android/armeabi-v7a)
endif()
# Android needs log
link_libraries(log)
elseif(CMAKE_SYSTEM_NAME STREQUAL "Linux")
link_directories(${NNSDK_ROOT}/linux/yocto/aarch64-poky-linux)
endif()
# Find OpenCV
message(STATUS "OpenCV_DIR: ${OpenCV_DIR}")
find_package(OpenCV REQUIRED)
include_directories(${OpenCV_INCLUDE_DIRS})
add_executable(paddleocr_det_demo
main.cpp
postprocess.cpp
postprocess.h
clipper.cpp
clipper.h
${CMAKE_SOURCE_DIR}/../../../../common/model_loader.cpp
)
target_link_libraries(paddleocr_det_demo
${OpenCV_LIBS}
nnsdk
)
cmake_minimum_required(VERSION 3.10...3.27)
project(yolo_world_demo)
set(CMAKE_CXX_STANDARD 17)
list(APPEND CMAKE_MODULE_PATH "${CMAKE_SOURCE_DIR}/../../../../cmake")
find_package(AMLNN REQUIRED)
include_directories(${AMLNN_INCLUDE_DIR})
link_directories(${AMLNN_LIBRARY_DIR})
include_directories(${CMAKE_SOURCE_DIR}/../../../../common)
# Set 3rdparty path
set(3RDPARTY_DIR "${CMAKE_SOURCE_DIR}/../../../../dependency")
if(CMAKE_SYSTEM_NAME STREQUAL "Android")
# Android needs log
link_libraries(log)
endif()
# Find OpenCV
message(STATUS "OpenCV_DIR: ${OpenCV_DIR}")
find_package(OpenCV REQUIRED)
include_directories(${OpenCV_INCLUDE_DIRS})
add_executable(paddleocr_det_demo
main.cpp
postprocess.cpp
postprocess.h
clipper.cpp
clipper.h
${CMAKE_SOURCE_DIR}/../../../../common/model_loader.cpp
)
target_link_libraries(paddleocr_det_demo
${OpenCV_LIBS}
${AMLNN_LIBRARY}
)

342
examples/resnet/README.md
View file

@ -1,165 +1,179 @@
# resnet
## 1.Overview
## 2.Model Download
- **Open Source model**
- **Open Source projects:**
- **Export Model Step:**
- **Install ultralytics**
pip install torch==2.4.1
pip install torchvision==0.19.1
pip install ultralytics==8.3.0
- **Download weights**
- **Export Model**
```
```
- **Exported Model**
​ link to amlogic server( **onnx model or quantized tflite**)
## 3. Model Conversion
```
cd model
Usage: ./adla_covnert.sh model_path adla_tookkit_path target_platform
example
```
| Parameter | Discription |
| ----------------- | ------------------------------------------------------------ |
| model_path | onnx model path |
| adla_tookkit_path | path to adla_toolkit |
| target_platform | Specify target platform. for A311D2 : PRODUCT_PID0XA003. for S905X5: PRODUCT_PID0XA005 |
## 4. Demo Run
### CPP
#### 1. Compile
**Prerequisites:**
- Android NDK (r25e recommended)
- `ANDROID_NDK_PATH` environment variable set
**Build:**
```bash
# Build for arm64-v8a
cd examples/resnet/cpp
./build-android.sh -a arm64-v8a
```
The executable will be generated at `build/android/resnet_demo` (Note: executable name may vary, verify in build folder).
#### 2. Run
```bash
# Push executable to device
adb push build/android/resnet_demo /data/local/tmp/
adb push model/res2net50_int8_A311D2.adla /data/local/tmp/
adb push imgs /data/local/tmp/
adb push labels.txt /data/local/tmp/
# Run on device
adb shell
cd /data/local/tmp
chmod +x resnet_demo
export LD_LIBRARY_PATH=/vendor/lib64 or (/vendor/lib)
# Usage: ./resnet_demo <model_path> <image_dir> <labels.txt>
./resnet_demo res2net50_int8_A311D2.adla imgs/ labels.txt
```
**Note:** Replace `res2net50_int8_A311D2.adla` with your actual model file path.
### Python
**Prerequisites:**
- Python 3.10
- Required packages: `numpy`, `opencv-python`, `amlnnlite`
**Install dependencies:**
```bash
pip install numpy opencv-python amlnnlite-1.0.0-cp310-cp310-linux_aarch64.whl
```
**Run on device:**
```bash
python resnet.py \
--model-path ./res2net50_int8_A311D2.adla \
--image-dir ./imgs \
--labels labels.txt \
--run-cycles 1 \
--loglevel INFO
```
Argument Descriptions:
| Argument | Description |
| ----------------- | ------------------------------------------------------------ |
| --board-work-path | Work path on board, default is /data/local/tmp |
| --model-path | path to .adla model |
| --image-dir | Directory containing test images |
| --labels | Path to synset_words.txt or labels.txt |
| --run-cycles | Number of inference cycles, default is 1 |
| --loglevel | Logging level: DEBUG / INFO / WARNING / ERROR, default is WARNING |
The script will automatically process all image files (`.jpg`, `.jpeg`, `.png`, `.bmp`) in the current directory and save results to a `{model_name}_result` folder.
## 5.Results
**Performance Feedback**
By setting the loglevel to INFO, the program provides real-time performance metrics upon completion. The console log will display essential hardware and execution details, including:
- Hardware Information: System and ADLA library versions.
- Model Overview: Basic input/output configurations.
- NPU Metrics: Total inference time (latency) and total DRAM bandwidth consumption.
**Classification Output**
For each image, the program prints the Top-5 classification results with their respective scores:
```bash
============================================================
Processing image 1/1: dog.jpg
============================================================ Top-5 Results:
1: Pekinese score=9.851644
2: West Highland white terrier score=5.055449
3: Maltese dog score=4.796195
4: basenji score=3.111045
5: Scotch terrier score=2.786978 ============================================================
```
**Profiling Visualization**
After a successful run of the Python demo, a folder named after the model (e.g., `{model_name}`) will be generated in the script directory. This folder contains 5 HTML files that provide a visual and detailed breakdown of per-layer performance:
- `hard_op_chart.html` & `soft_op_chart.html`: Hardware/Software op execution details.
- `dram_rd_chart.html` & `dram_wr_chart.html`: Bandwidth read/write distribution.
- `pie_charts_distribution.html`: Overall resource allocation.
You can pull the result folder back to view it:
```bash
adb pull /data/local/tmp/res2net50_int8_A311D2
```
Taking hard_op_chart.html as an example (shown below), each layer's ADLA operator name includes parentheses containing the index of the corresponding quantized .tflite layer(s); by default, these indices are suppressed, and operators are labeled generically as "hardware" or "software" without numerical suffixes.
# resnet
## 1.Overview
## 2.Model Download
- **Open Source model**
- **Open Source projects:**
- **Export Model Step:**
- **Install ultralytics**
pip install torch==2.4.1
pip install torchvision==0.19.1
pip install ultralytics==8.3.0
- **Download weights**
- **Export Model**
```
```
- **Exported Model**
​ link to amlogic server( **onnx model or quantized tflite**)
## 3. Model Conversion
```
cd model
Usage: ./adla_covnert.sh model_path adla_tookkit_path target_platform
example
```
| Parameter | Discription |
| ----------------- | ------------------------------------------------------------ |
| model_path | onnx model path |
| adla_tookkit_path | path to adla_toolkit |
| target_platform | Specify target platform. for A311D2 : PRODUCT_PID0XA003. for S905X5: PRODUCT_PID0XA005 |
## 4. Demo Run
### CPP
#### 1. Compile
#### AMLNN SDK Setup
Resolve the AMLNN nnsdk dependency using one of the following methods:
- **Priority 1 – Environment variable (recommended)**
```bash
export AMLNN_HOME=/path/to/amlnn-toolkit
```
- **Priority 3 – Sibling directory fallback** *(automatic)*
Place `amlnn-toolkit` as a sibling to `amlnn-model-playground`:
```bash
git clone git@github.com:Amlogic-NN/amlnn-toolkit.git ../amlnn-toolkit
```
**Prerequisites:**
- Android NDK (r25e recommended)
- `ANDROID_NDK_PATH` environment variable set
**Build:**
```bash
# Build for arm64-v8a
cd examples/resnet/cpp
./build-android.sh -a arm64-v8a
```
The executable will be generated at `build/android/resnet_demo` (Note: executable name may vary, verify in build folder).
#### 2. Run
```bash
# Push executable to device
adb push build/android/resnet_demo /data/local/tmp/
adb push model/res2net50_int8_A311D2.adla /data/local/tmp/
adb push imgs /data/local/tmp/
adb push labels.txt /data/local/tmp/
# Run on device
adb shell
cd /data/local/tmp
chmod +x resnet_demo
export LD_LIBRARY_PATH=/vendor/lib64 or (/vendor/lib)
# Usage: ./resnet_demo <model_path> <image_dir> <labels.txt>
./resnet_demo res2net50_int8_A311D2.adla imgs/ labels.txt
```
**Note:** Replace `res2net50_int8_A311D2.adla` with your actual model file path.
### Python
**Prerequisites:**
- Python 3.10
- Required packages: `numpy`, `opencv-python`, `amlnnlite`
**Install dependencies:**
```bash
pip install numpy opencv-python amlnnlite-1.0.0-cp310-cp310-linux_aarch64.whl
```
**Run on device:**
```bash
python resnet.py \
--model-path ./res2net50_int8_A311D2.adla \
--image-dir ./imgs \
--labels labels.txt \
--run-cycles 1 \
--loglevel INFO
```
Argument Descriptions:
| Argument | Description |
| ----------------- | ------------------------------------------------------------ |
| --board-work-path | Work path on board, default is /data/local/tmp |
| --model-path | path to .adla model |
| --image-dir | Directory containing test images |
| --labels | Path to synset_words.txt or labels.txt |
| --run-cycles | Number of inference cycles, default is 1 |
| --loglevel | Logging level: DEBUG / INFO / WARNING / ERROR, default is WARNING |
The script will automatically process all image files (`.jpg`, `.jpeg`, `.png`, `.bmp`) in the current directory and save results to a `{model_name}_result` folder.
## 5.Results
**Performance Feedback**
By setting the loglevel to INFO, the program provides real-time performance metrics upon completion. The console log will display essential hardware and execution details, including:
- Hardware Information: System and ADLA library versions.
- Model Overview: Basic input/output configurations.
- NPU Metrics: Total inference time (latency) and total DRAM bandwidth consumption.
**Classification Output**
For each image, the program prints the Top-5 classification results with their respective scores:
```bash
============================================================
Processing image 1/1: dog.jpg
============================================================ Top-5 Results:
1: Pekinese score=9.851644
2: West Highland white terrier score=5.055449
3: Maltese dog score=4.796195
4: basenji score=3.111045
5: Scotch terrier score=2.786978 ============================================================
```
**Profiling Visualization**
After a successful run of the Python demo, a folder named after the model (e.g., `{model_name}`) will be generated in the script directory. This folder contains 5 HTML files that provide a visual and detailed breakdown of per-layer performance:
- `hard_op_chart.html` & `soft_op_chart.html`: Hardware/Software op execution details.
- `dram_rd_chart.html` & `dram_wr_chart.html`: Bandwidth read/write distribution.
- `pie_charts_distribution.html`: Overall resource allocation.
You can pull the result folder back to view it:
```bash
adb pull /data/local/tmp/res2net50_int8_A311D2
```
Taking hard_op_chart.html as an example (shown below), each layer's ADLA operator name includes parentheses containing the index of the corresponding quantized .tflite layer(s); by default, these indices are suppressed, and operators are labeled generically as "hardware" or "software" without numerical suffixes.
![alt text](Visualization.png)

3
examples/resnet/cpp/build-android.sh
View file

@ -65,7 +65,8 @@ echo "BUILD_DIR: ${BUILD_DIR}"
mkdir -p ${BUILD_DIR}
cd ${BUILD_DIR}
cmake ../../src \
cmake -Wno-dev ../../src \
-DAMLNN_HOME=${AMLNN_HOME:-} \
-DCMAKE_TOOLCHAIN_FILE=${ANDROID_NDK_PATH}/build/cmake/android.toolchain.cmake \
-DANDROID_ABI=${TARGET_ABI} \
-DANDROID_PLATFORM=android-24 \

79
examples/resnet/cpp/src/CMakeLists.txt
View file

@ -1,46 +1,35 @@
cmake_minimum_required(VERSION 3.5)
project(resnet_demo)
set(CMAKE_CXX_STANDARD 17)
# Set NNSDK path
if(NOT DEFINED NNSDK_DIR)
set(NNSDK_DIR "${CMAKE_SOURCE_DIR}/../../../../../amlnn-toolkit/nn_runtime/nnsdk")
endif()
set(NNSDK_ROOT "${NNSDK_DIR}")
message(STATUS "NNSDK_ROOT: ${NNSDK_ROOT}")
include_directories(${NNSDK_ROOT}/include)
include_directories(${CMAKE_SOURCE_DIR}/../../../../common)
# Set dependency path
set(3RDPARTY_DIR "${CMAKE_SOURCE_DIR}/../../../../dependency")
if(CMAKE_SYSTEM_NAME STREQUAL "Android")
if (ANDROID_ABI STREQUAL "arm64-v8a")
link_directories(${NNSDK_ROOT}/android/arm64-v8a)
else()
link_directories(${NNSDK_ROOT}/android/armeabi-v7a)
endif()
# Android needs log
link_libraries(log)
elseif(CMAKE_SYSTEM_NAME STREQUAL "Linux")
link_directories(${NNSDK_ROOT}/linux/yocto/aarch64-poky-linux)
endif()
# Find OpenCV
message(STATUS "OpenCV_DIR: ${OpenCV_DIR}")
find_package(OpenCV REQUIRED)
include_directories(${OpenCV_INCLUDE_DIRS})
add_executable(resnet_demo
main.cpp
postprocess.cpp
${CMAKE_SOURCE_DIR}/../../../../common/model_loader.cpp
)
target_link_libraries(resnet_demo
${OpenCV_LIBS}
nnsdk
cmake_minimum_required(VERSION 3.10...3.27)
project(resnet_demo)
set(CMAKE_CXX_STANDARD 17)
list(APPEND CMAKE_MODULE_PATH "${CMAKE_SOURCE_DIR}/../../../../cmake")
find_package(AMLNN REQUIRED)
include_directories(${AMLNN_INCLUDE_DIR})
link_directories(${AMLNN_LIBRARY_DIR})
include_directories(${CMAKE_SOURCE_DIR}/../../../../common)
# Set dependency path
set(3RDPARTY_DIR "${CMAKE_SOURCE_DIR}/../../../../dependency")
if(CMAKE_SYSTEM_NAME STREQUAL "Android")
# Android needs log
link_libraries(log)
endif()
# Find OpenCV
message(STATUS "OpenCV_DIR: ${OpenCV_DIR}")
find_package(OpenCV REQUIRED)
include_directories(${OpenCV_INCLUDE_DIRS})
add_executable(resnet_demo
main.cpp
postprocess.cpp
${CMAKE_SOURCE_DIR}/../../../../common/model_loader.cpp
)
target_link_libraries(resnet_demo
${OpenCV_LIBS}
${AMLNN_LIBRARY}
)

334
examples/retinaface/README.md
View file

@ -1,160 +1,174 @@
# retinaface
## 1.Overview
## 2.Model Download
- **Open Source model**
- **Open Source projects:**
- **Export Model Step:**
- **Install ultralytics**
pip install torch==2.4.1
pip install torchvision==0.19.1
pip install ultralytics==8.3.0
- **Download weights**
- **Export Model**
- **Exported Model**
​ link to amlogic server( **onnx model or quantized tflite**)
## 3. Model Conversion
```
cd model
Usage: ./adla_covnert.sh model_path adla_tookkit_path target_platform
example
```
| Parameter | Discription |
| ----------------- | ------------------------------------------------------------ |
| model_path | onnx model path |
| adla_tookkit_path | path to adla_toolkit |
| target_platform | Specify target platform. for A311D2 : PRODUCT_PID0XA003. for S905X5: PRODUCT_PID0XA005 |
## 4. Demo Run
### CPP
#### 1. Compile
**Prerequisites:**
- Android NDK (r25e recommended)
- `ANDROID_NDK_PATH` environment variable set
**Build:**
```bash
# Build for arm64-v8a
cd examples/retinaface/cpp
./build-android.sh -a arm64-v8a
```
The executable will be generated at `build/android/retinaface_demo` (Note: executable name may vary, verify in build folder).
#### 2. Run
```bash
# Push executable to device
adb push build/android/retinaface_demo /data/local/tmp/
adb push model/RetinaFace_int8_A311D2.adla /data/local/tmp/
adb push imgs /data/local/tmp/
# Run on device
adb shell
cd /data/local/tmp
chmod +x retinaface_demo
export LD_LIBRARY_PATH=/vendor/lib64 or (/vendor/lib)
# Usage: ./retinaface_demo <model_path> <image_dir>
./retinaface_demo RetinaFace_int8_A311D2.adla ./imgs
```
**Note:** Replace `RetinaFace_int8_A311D2.adla` with your actual model file path.
### Python
**Prerequisites:**
- Python 3.10
- Required packages: `numpy`, `opencv-python`, `amlnnlite`
**Install dependencies:**
```bash
pip install numpy opencv-python amlnnlite-1.0.0-cp310-cp310-linux_aarch64.whl
```
**Run on device:**
```bash
python RetinaFace.py \
--model-path ./RetinaFace_int8_A311D2.adla \
--image-dir ./imgs \
--run-cycles 1 \
--loglevel INFO
```
Argument Descriptions:
| Argument | Description |
| ----------------- | ------------------------------------------------------------ |
| --board-work-path | Work path on board, default is /data/local/tmp |
| --model-path | path to .adla model |
| --image-dir | Directory containing test images |
| --run-cycles | Number of inference cycles, default is 1 |
| --loglevel | Logging level: DEBUG / INFO / WARNING / ERROR, default is WARNING |
The script will automatically process all image files (`.jpg`, `.jpeg`, `.png`, `.bmp`) in the current directory and save results to a `{model_name}_result` folder.
## 5.Results
**Performance Feedback**
By setting the loglevel to INFO, the program provides real-time performance metrics upon completion. The console log will display essential hardware and execution details, including:
- Hardware Information: System and ADLA library versions.
- Model Overview: Basic input/output configurations.
- NPU Metrics: Total inference time (latency) and total DRAM bandwidth consumption.
**Detection Output**
The program will print the detection count. The output images, featuring bounding boxes and five facial landmarks (eyes, nose, and mouth corners), will be saved to the `{model_name}_result` folder.
You can pull the result folder back to view it:
```bash
adb pull /data/local/tmp/RetinaFace_int8_A311D2_result
```
![alt text](result.jpg)
**Profiling Visualization**
After a successful run of the Python demo, a folder named after the model (e.g., `{model_name}`) will be generated in the script directory. This folder contains 5 HTML files that provide a visual and detailed breakdown of per-layer performance:
- `hard_op_chart.html` & `soft_op_chart.html`: Hardware/Software op execution details.
- `dram_rd_chart.html` & `dram_wr_chart.html`: Bandwidth read/write distribution.
- `pie_charts_distribution.html`: Overall resource allocation.
You can pull the result folder back to view it:
```bash
adb pull /data/local/tmp/RetinaFace_int8_A311D2
```
Taking hard_op_chart.html as an example (shown below), each layer's ADLA operator name includes parentheses containing the index of the corresponding quantized .tflite layer(s); by default, these indices are suppressed, and operators are labeled generically as "hardware" or "software" without numerical suffixes.
![alt text](Visualization.png)
# retinaface
## 1.Overview
## 2.Model Download
- **Open Source model**
- **Open Source projects:**
- **Export Model Step:**
- **Install ultralytics**
pip install torch==2.4.1
pip install torchvision==0.19.1
pip install ultralytics==8.3.0
- **Download weights**
- **Export Model**
- **Exported Model**
​ link to amlogic server( **onnx model or quantized tflite**)
## 3. Model Conversion
```
cd model
Usage: ./adla_covnert.sh model_path adla_tookkit_path target_platform
example
```
| Parameter | Discription |
| ----------------- | ------------------------------------------------------------ |
| model_path | onnx model path |
| adla_tookkit_path | path to adla_toolkit |
| target_platform | Specify target platform. for A311D2 : PRODUCT_PID0XA003. for S905X5: PRODUCT_PID0XA005 |
## 4. Demo Run
### CPP
#### 1. Compile
#### AMLNN SDK Setup
Resolve the AMLNN nnsdk dependency using one of the following methods:
- **Priority 1 – Environment variable (recommended)**
```bash
export AMLNN_HOME=/path/to/amlnn-toolkit
```
- **Priority 3 – Sibling directory fallback** *(automatic)*
Place `amlnn-toolkit` as a sibling to `amlnn-model-playground`:
```bash
git clone git@github.com:Amlogic-NN/amlnn-toolkit.git ../amlnn-toolkit
```
**Prerequisites:**
- Android NDK (r25e recommended)
- `ANDROID_NDK_PATH` environment variable set
**Build:**
```bash
# Build for arm64-v8a
cd examples/retinaface/cpp
./build-android.sh -a arm64-v8a
```
The executable will be generated at `build/android/retinaface_demo` (Note: executable name may vary, verify in build folder).
#### 2. Run
```bash
# Push executable to device
adb push build/android/retinaface_demo /data/local/tmp/
adb push model/RetinaFace_int8_A311D2.adla /data/local/tmp/
adb push imgs /data/local/tmp/
# Run on device
adb shell
cd /data/local/tmp
chmod +x retinaface_demo
export LD_LIBRARY_PATH=/vendor/lib64 or (/vendor/lib)
# Usage: ./retinaface_demo <model_path> <image_dir>
./retinaface_demo RetinaFace_int8_A311D2.adla ./imgs
```
**Note:** Replace `RetinaFace_int8_A311D2.adla` with your actual model file path.
### Python
**Prerequisites:**
- Python 3.10
- Required packages: `numpy`, `opencv-python`, `amlnnlite`
**Install dependencies:**
```bash
pip install numpy opencv-python amlnnlite-1.0.0-cp310-cp310-linux_aarch64.whl
```
**Run on device:**
```bash
python RetinaFace.py \
--model-path ./RetinaFace_int8_A311D2.adla \
--image-dir ./imgs \
--run-cycles 1 \
--loglevel INFO
```
Argument Descriptions:
| Argument | Description |
| ----------------- | ------------------------------------------------------------ |
| --board-work-path | Work path on board, default is /data/local/tmp |
| --model-path | path to .adla model |
| --image-dir | Directory containing test images |
| --run-cycles | Number of inference cycles, default is 1 |
| --loglevel | Logging level: DEBUG / INFO / WARNING / ERROR, default is WARNING |
The script will automatically process all image files (`.jpg`, `.jpeg`, `.png`, `.bmp`) in the current directory and save results to a `{model_name}_result` folder.
## 5.Results
**Performance Feedback**
By setting the loglevel to INFO, the program provides real-time performance metrics upon completion. The console log will display essential hardware and execution details, including:
- Hardware Information: System and ADLA library versions.
- Model Overview: Basic input/output configurations.
- NPU Metrics: Total inference time (latency) and total DRAM bandwidth consumption.
**Detection Output**
The program will print the detection count. The output images, featuring bounding boxes and five facial landmarks (eyes, nose, and mouth corners), will be saved to the `{model_name}_result` folder.
You can pull the result folder back to view it:
```bash
adb pull /data/local/tmp/RetinaFace_int8_A311D2_result
```
![alt text](result.jpg)
**Profiling Visualization**
After a successful run of the Python demo, a folder named after the model (e.g., `{model_name}`) will be generated in the script directory. This folder contains 5 HTML files that provide a visual and detailed breakdown of per-layer performance:
- `hard_op_chart.html` & `soft_op_chart.html`: Hardware/Software op execution details.
- `dram_rd_chart.html` & `dram_wr_chart.html`: Bandwidth read/write distribution.
- `pie_charts_distribution.html`: Overall resource allocation.
You can pull the result folder back to view it:
```bash
adb pull /data/local/tmp/RetinaFace_int8_A311D2
```
Taking hard_op_chart.html as an example (shown below), each layer's ADLA operator name includes parentheses containing the index of the corresponding quantized .tflite layer(s); by default, these indices are suppressed, and operators are labeled generically as "hardware" or "software" without numerical suffixes.
![alt text](Visualization.png)

3
examples/retinaface/cpp/build-android.sh
View file

@ -65,7 +65,8 @@ echo "BUILD_DIR: ${BUILD_DIR}"
mkdir -p ${BUILD_DIR}
cd ${BUILD_DIR}
cmake ../../src \
cmake -Wno-dev ../../src \
-DAMLNN_HOME=${AMLNN_HOME:-} \
-DCMAKE_TOOLCHAIN_FILE=${ANDROID_NDK_PATH}/build/cmake/android.toolchain.cmake \
-DANDROID_ABI=${TARGET_ABI} \
-DANDROID_PLATFORM=android-24 \

79
examples/retinaface/cpp/src/CMakeLists.txt
View file

@ -1,46 +1,35 @@
cmake_minimum_required(VERSION 3.5)
project(retinaface_demo)
set(CMAKE_CXX_STANDARD 17)
# Set NNSDK path
if(NOT DEFINED NNSDK_DIR)
set(NNSDK_DIR "${CMAKE_SOURCE_DIR}/../../../../../amlnn-toolkit/nn_runtime/nnsdk")
endif()
set(NNSDK_ROOT "${NNSDK_DIR}")
message(STATUS "NNSDK_ROOT: ${NNSDK_ROOT}")
include_directories(${NNSDK_ROOT}/include)
include_directories(${CMAKE_SOURCE_DIR}/../../../../common)
# Set dependency path
set(3RDPARTY_DIR "${CMAKE_SOURCE_DIR}/../../../../dependency")
if(CMAKE_SYSTEM_NAME STREQUAL "Android")
if (ANDROID_ABI STREQUAL "arm64-v8a")
link_directories(${NNSDK_ROOT}/android/arm64-v8a)
else()
link_directories(${NNSDK_ROOT}/android/armeabi-v7a)
endif()
# Android needs log
link_libraries(log)
elseif(CMAKE_SYSTEM_NAME STREQUAL "Linux")
link_directories(${NNSDK_ROOT}/linux/yocto/aarch64-poky-linux)
endif()
# Find OpenCV
message(STATUS "OpenCV_DIR: ${OpenCV_DIR}")
find_package(OpenCV REQUIRED)
include_directories(${OpenCV_INCLUDE_DIRS})
add_executable(retinaface_demo
main.cpp
postprocess.cpp
${CMAKE_SOURCE_DIR}/../../../../common/model_loader.cpp
)
target_link_libraries(retinaface_demo
${OpenCV_LIBS}
nnsdk
cmake_minimum_required(VERSION 3.10...3.27)
project(retinaface_demo)
set(CMAKE_CXX_STANDARD 17)
list(APPEND CMAKE_MODULE_PATH "${CMAKE_SOURCE_DIR}/../../../../cmake")
find_package(AMLNN REQUIRED)
include_directories(${AMLNN_INCLUDE_DIR})
link_directories(${AMLNN_LIBRARY_DIR})
include_directories(${CMAKE_SOURCE_DIR}/../../../../common)
# Set dependency path
set(3RDPARTY_DIR "${CMAKE_SOURCE_DIR}/../../../../dependency")
if(CMAKE_SYSTEM_NAME STREQUAL "Android")
# Android needs log
link_libraries(log)
endif()
# Find OpenCV
message(STATUS "OpenCV_DIR: ${OpenCV_DIR}")
find_package(OpenCV REQUIRED)
include_directories(${OpenCV_INCLUDE_DIRS})
add_executable(retinaface_demo
main.cpp
postprocess.cpp
${CMAKE_SOURCE_DIR}/../../../../common/model_loader.cpp
)
target_link_libraries(retinaface_demo
${OpenCV_LIBS}
${AMLNN_LIBRARY}
)

34
examples/whisper/README.md
View file

@ -0,0 +1,34 @@
# Whisper
## 4. Demo Run
### CPP
#### 1. Compile
#### AMLNN SDK Setup
Resolve the AMLNN nnsdk dependency using one of the following methods:
- **Priority 1 – Environment variable (recommended)**
```bash
export AMLNN_HOME=/path/to/amlnn-toolkit
```
- **Priority 3 – Sibling directory fallback** *(automatic)*
Place `amlnn-toolkit` as a sibling to `amlnn-model-playground`:
```bash
git clone git@github.com:Amlogic-NN/amlnn-toolkit.git ../amlnn-toolkit
```
**Prerequisites:**
- Android NDK r25c
- `ANDROID_NDK_PATH` environment variable set
**Build:**
```bash
# Build for arm64-v8a
cd examples/whisper/cpp
./build-android.sh -a arm64-v8a
```
The executable will be generated in `build/android/`.

3
examples/whisper/cpp/build-android.sh
View file

@ -65,7 +65,8 @@ echo "BUILD_DIR: ${BUILD_DIR}"
mkdir -p ${BUILD_DIR}
cd ${BUILD_DIR}
cmake ../../src \
cmake -Wno-dev ../../src \
-DAMLNN_HOME=${AMLNN_HOME:-} \
-DCMAKE_TOOLCHAIN_FILE=${ANDROID_NDK_PATH}/build/cmake/android.toolchain.cmake \
-DANDROID_ABI=${TARGET_ABI} \
-DANDROID_PLATFORM=android-24 \

80
examples/whisper/cpp/src/CMakeLists.txt
View file

@ -1,45 +1,35 @@
cmake_minimum_required(VERSION 3.5)
project(whisper_demo)
set(CMAKE_CXX_STANDARD 17)
# Set NNSDK path
if(NOT DEFINED NNSDK_DIR)
set(NNSDK_DIR "${CMAKE_SOURCE_DIR}/../../../../../amlnn-toolkit/nn_runtime/nnsdk")
endif()
set(NNSDK_ROOT "${NNSDK_DIR}")
message(STATUS "NNSDK_ROOT: ${NNSDK_ROOT}")
include_directories(${NNSDK_ROOT}/include)
include_directories(${CMAKE_SOURCE_DIR}/../../../../common)
# Set 3rdparty path
set(3RDPARTY_DIR "${CMAKE_SOURCE_DIR}/../../../../dependency")
if(CMAKE_SYSTEM_NAME STREQUAL "Android")
if (ANDROID_ABI STREQUAL "arm64-v8a")
link_directories(${NNSDK_ROOT}/android/arm64-v8a)
else()
link_directories(${NNSDK_ROOT}/android/armeabi-v7a)
endif()
# Android needs log
link_libraries(log)
elseif(CMAKE_SYSTEM_NAME STREQUAL "Linux")
link_directories(${NNSDK_ROOT}/linux/yocto/aarch64-poky-linux)
endif()
add_executable(${PROJECT_NAME}
main.cpp
common.cpp
whisper.cpp
whisper_invoke.cpp
pre_process_whisper.cpp
post_process_whisper.cpp
)
target_link_libraries(${PROJECT_NAME}
nnsdk
dl
m
)
cmake_minimum_required(VERSION 3.10...3.27)
project(whisper_demo)
set(CMAKE_CXX_STANDARD 17)
list(APPEND CMAKE_MODULE_PATH "${CMAKE_SOURCE_DIR}/../../../../cmake")
find_package(AMLNN REQUIRED)
include_directories(${AMLNN_INCLUDE_DIR})
link_directories(${AMLNN_LIBRARY_DIR})
include_directories(${CMAKE_SOURCE_DIR}/../../../../common)
# Set 3rdparty path
set(3RDPARTY_DIR "${CMAKE_SOURCE_DIR}/../../../../dependency")
if(CMAKE_SYSTEM_NAME STREQUAL "Android")
# Android needs log
link_libraries(log)
endif()
add_executable(${PROJECT_NAME}
main.cpp
common.cpp
whisper.cpp
whisper_invoke.cpp
pre_process_whisper.cpp
post_process_whisper.cpp
)
target_link_libraries(${PROJECT_NAME}
${AMLNN_LIBRARY}
dl
m
)

34
examples/yoloe/README.md
View file

@ -0,0 +1,34 @@
# YOLOE
## 4. Demo Run
### CPP
#### 1. Compile
#### AMLNN SDK Setup
Resolve the AMLNN nnsdk dependency using one of the following methods:
- **Priority 1 – Environment variable (recommended)**
```bash
export AMLNN_HOME=/path/to/amlnn-toolkit
```
- **Priority 3 – Sibling directory fallback** *(automatic)*
Place `amlnn-toolkit` as a sibling to `amlnn-model-playground`:
```bash
git clone git@github.com:Amlogic-NN/amlnn-toolkit.git ../amlnn-toolkit
```
**Prerequisites:**
- Android NDK r25c
- `ANDROID_NDK_PATH` environment variable set
**Build:**
```bash
# Build for arm64-v8a
cd examples/yoloe/cpp
./build-android.sh -a arm64-v8a
```
The executable will be generated in `build/android/`.

3
examples/yoloe/cpp/build-android.sh
View file

@ -65,7 +65,8 @@ echo "BUILD_DIR: ${BUILD_DIR}"
mkdir -p ${BUILD_DIR}
cd ${BUILD_DIR}
cmake ../../src \
cmake -Wno-dev ../../src \
-DAMLNN_HOME=${AMLNN_HOME:-} \
-DCMAKE_TOOLCHAIN_FILE=${ANDROID_NDK_PATH}/build/cmake/android.toolchain.cmake \
-DANDROID_ABI=${TARGET_ABI} \
-DANDROID_PLATFORM=android-24 \

22
examples/yoloe/cpp/src/CMakeLists.txt
View file

@ -1,31 +1,21 @@
cmake_minimum_required(VERSION 3.5)
cmake_minimum_required(VERSION 3.10...3.27)
project(yoloe_demo)
set(CMAKE_CXX_STANDARD 17)
# Set NNSDK path
if(NOT DEFINED NNSDK_DIR)
set(NNSDK_DIR "${CMAKE_SOURCE_DIR}/../../../../../amlnn-toolkit/nn_runtime/nnsdk")
endif()
set(NNSDK_ROOT "${NNSDK_DIR}")
message(STATUS "NNSDK_ROOT: ${NNSDK_ROOT}")
list(APPEND CMAKE_MODULE_PATH "${CMAKE_SOURCE_DIR}/../../../../cmake")
find_package(AMLNN REQUIRED)
include_directories(${AMLNN_INCLUDE_DIR})
link_directories(${AMLNN_LIBRARY_DIR})
include_directories(${NNSDK_ROOT}/include)
include_directories(${CMAKE_SOURCE_DIR}/../../../../common)
# Set dependency path
set(3RDPARTY_DIR "${CMAKE_SOURCE_DIR}/../../../../dependency")
if(CMAKE_SYSTEM_NAME STREQUAL "Android")
if (ANDROID_ABI STREQUAL "arm64-v8a")
link_directories(${NNSDK_ROOT}/android/arm64-v8a)
else()
link_directories(${NNSDK_ROOT}/android/armeabi-v7a)
endif()
# Android needs log
link_libraries(log)
elseif(CMAKE_SYSTEM_NAME STREQUAL "Linux")
link_directories(${NNSDK_ROOT}/linux/yocto/aarch64-poky-linux)
endif()
# Find OpenCV
@ -40,5 +30,5 @@ add_executable(yoloe_demo
target_link_libraries(yoloe_demo
${OpenCV_LIBS}
nnsdk
${AMLNN_LIBRARY}
)

352
examples/yolov11/README.md
View file

@ -1,170 +1,184 @@
# yolov11
## 1.Overview
​ YOLOv11 was released by Ultralytics on October 2, 2024. It introduces significant architectural refinements and efficiency optimizations, delivering superior accuracy-speed trade-offs compared to previous generations. With enhanced feature extraction capabilities, YOLOv11 is designed for high-performance real-time applications—including object detection, instance segmentation, and pose estimation—to handle demanding tasks in a wide range of applications.
## 2.Model Download
- **Open Source model**
- **Open Source projects:** https://github.com/ultralytics/ultralytics/tree/v8.3.0
- **Export Model Step:**
- **Install ultralytics**
pip install torch==2.4.1
pip install torchvision==0.19.1
pip install ultralytics==8.3.0
- **Download weights**
wget https://github.com/ultralytics/assets/releases/download/v8.3.0/yolo11m.pt
wget https://github.com/ultralytics/assets/releases/download/v8.3.0/yolo11s.pt
wget https://github.com/ultralytics/assets/releases/download/v8.3.0/yolo11n.pt
- **Export Model**
```
from ultralytics import YOLO
model = YOLO("yolo11n.pt")
model.export(format="onnx", opset=12, simplify=True, dynamic=False, imgsz=640)
```
- **Exported Model**
​ link to amlogic server( **onnx model or quantized tflite**)
## 3. Model Conversion
```
cd model
Usage: ./adla_covnert.sh model_path adla_tookkit_path target_platform
example
./adla_covnert.sh yolov11m.onnx /xxxx/adla-toolkit-binary-3.2.9.3 PRODUCT_PID0XA005
./adla_covnert.sh yolov11s.onnx /xxxx/adla-toolkit-binary-3.2.9.3 PRODUCT_PID0XA005
./adla_covnert.sh yolov11n.onnx /xxxx/adla-toolkit-binary-3.2.9.3 PRODUCT_PID0XA005
```
| Parameter | Discription |
| ----------------- | ------------------------------------------------------------ |
| model_path | onnx model path |
| adla_tookkit_path | path to adla_toolkit |
| target_platform | Specify target platform. for A311D2 : PRODUCT_PID0XA003. for S905X5: PRODUCT_PID0XA005 |
## 4. Demo Run
### CPP
#### 1. Compile
**Prerequisites:**
- Android NDK (r25e recommended)
- `ANDROID_NDK_PATH` environment variable set
**Build:**
```bash
# Build for arm64-v8a
cd examples/yolov11/cpp
./build-android.sh -a arm64-v8a
```
The executable will be generated at `build/android/yolo11_demo` (Note: executable name may vary, verify in build folder).
#### 2. Run
```bash
# Push executable to device
adb push build/android/yolo11_demo /data/local/tmp/
adb push model/yolov11n_int8_A311D2.adla /data/local/tmp/
adb push imgs /data/local/tmp/
# Run on device
adb shell
cd /data/local/tmp
chmod +x yolo11_demo
export LD_LIBRARY_PATH=/vendor/lib64 or (/vendor/lib)
# Usage: ./yolo11_demo <model_path> <image_dir>
./yolo11_demo yolov11n_int8_A311D2.adla ./imgs
```
**Note:** Replace `yolov11n_int8_A311D2.adla` with your actual model file path.
### Python
**Prerequisites:**
- Python 3.10
- Required packages: `numpy`, `opencv-python`, `amlnnlite`
**Install dependencies:**
```bash
pip install numpy opencv-python amlnnlite-1.0.0-cp310-cp310-linux_aarch64.whl
```
**Run on device:**
```bash
python yolov11.py \
--model-path ./yolov11n_int8_A311D2.adla \
--image-dir ./imgs \
--run-cycles 1 \
--loglevel INFO
```
Argument Descriptions:
| Argument | Description |
| ----------------- | ------------------------------------------------------------ |
| --board-work-path | Work path on board, default is /data/local/tmp |
| --model-path | path to .adla model |
| --image-dir | Directory containing test images |
| --run-cycles | Number of inference cycles, default is 1 |
| --loglevel | Logging level: DEBUG / INFO / WARNING / ERROR, default is WARNING |
The script will automatically process all image files (`.jpg`, `.jpeg`, `.png`, `.bmp`) in the current directory and save results to a `{model_name}_result` folder.
## 5.Results
**Performance Feedback**
By setting the loglevel to INFO, the program provides real-time performance metrics upon completion. The console log will display essential hardware and execution details, including:
- Hardware Information: System and ADLA library versions.
- Model Overview: Basic input/output configurations.
- NPU Metrics: Total inference time (latency) and total DRAM bandwidth consumption.
**Detection Output**
The program will print the detection count. The result image with bounding boxes will be saved to the specified output path (`{model_name}_result`).
You can pull the result folder back to view it:
```bash
adb pull /data/local/tmp/yolov11n_int8_A311D2_result
```
![alt text](result.jpg)
**Profiling Visualization**
After a successful run of the Python demo, a folder named after the model (e.g., `{model_name}`) will be generated in the script directory. This folder contains 5 HTML files that provide a visual and detailed breakdown of per-layer performance:
- `hard_op_chart.html` & `soft_op_chart.html`: Hardware/Software op execution details.
- `dram_rd_chart.html` & `dram_wr_chart.html`: Bandwidth read/write distribution.
- `pie_charts_distribution.html`: Overall resource allocation.
You can pull the result folder back to view it:
```bash
adb pull /data/local/tmp/yolov11n_int8_A311D2
```
Taking hard_op_chart.html as an example (shown below), each layer's ADLA operator name includes parentheses containing the index of the corresponding quantized .tflite layer(s); by default, these indices are suppressed, and operators are labeled generically as "hardware" or "software" without numerical suffixes.
# yolov11
## 1.Overview
​ YOLOv11 was released by Ultralytics on October 2, 2024. It introduces significant architectural refinements and efficiency optimizations, delivering superior accuracy-speed trade-offs compared to previous generations. With enhanced feature extraction capabilities, YOLOv11 is designed for high-performance real-time applications—including object detection, instance segmentation, and pose estimation—to handle demanding tasks in a wide range of applications.
## 2.Model Download
- **Open Source model**
- **Open Source projects:** https://github.com/ultralytics/ultralytics/tree/v8.3.0
- **Export Model Step:**
- **Install ultralytics**
pip install torch==2.4.1
pip install torchvision==0.19.1
pip install ultralytics==8.3.0
- **Download weights**
wget https://github.com/ultralytics/assets/releases/download/v8.3.0/yolo11m.pt
wget https://github.com/ultralytics/assets/releases/download/v8.3.0/yolo11s.pt
wget https://github.com/ultralytics/assets/releases/download/v8.3.0/yolo11n.pt
- **Export Model**
```
from ultralytics import YOLO
model = YOLO("yolo11n.pt")
model.export(format="onnx", opset=12, simplify=True, dynamic=False, imgsz=640)
```
- **Exported Model**
​ link to amlogic server( **onnx model or quantized tflite**)
## 3. Model Conversion
```
cd model
Usage: ./adla_covnert.sh model_path adla_tookkit_path target_platform
example
./adla_covnert.sh yolov11m.onnx /xxxx/adla-toolkit-binary-3.2.9.3 PRODUCT_PID0XA005
./adla_covnert.sh yolov11s.onnx /xxxx/adla-toolkit-binary-3.2.9.3 PRODUCT_PID0XA005
./adla_covnert.sh yolov11n.onnx /xxxx/adla-toolkit-binary-3.2.9.3 PRODUCT_PID0XA005
```
| Parameter | Discription |
| ----------------- | ------------------------------------------------------------ |
| model_path | onnx model path |
| adla_tookkit_path | path to adla_toolkit |
| target_platform | Specify target platform. for A311D2 : PRODUCT_PID0XA003. for S905X5: PRODUCT_PID0XA005 |
## 4. Demo Run
### CPP
#### 1. Compile
#### AMLNN SDK Setup
Resolve the AMLNN nnsdk dependency using one of the following methods:
- **Priority 1 – Environment variable (recommended)**
```bash
export AMLNN_HOME=/path/to/amlnn-toolkit
```
- **Priority 3 – Sibling directory fallback** *(automatic)*
Place `amlnn-toolkit` as a sibling to `amlnn-model-playground`:
```bash
git clone git@github.com:Amlogic-NN/amlnn-toolkit.git ../amlnn-toolkit
```
**Prerequisites:**
- Android NDK (r25e recommended)
- `ANDROID_NDK_PATH` environment variable set
**Build:**
```bash
# Build for arm64-v8a
cd examples/yolov11/cpp
./build-android.sh -a arm64-v8a
```
The executable will be generated at `build/android/yolo11_demo` (Note: executable name may vary, verify in build folder).
#### 2. Run
```bash
# Push executable to device
adb push build/android/yolo11_demo /data/local/tmp/
adb push model/yolov11n_int8_A311D2.adla /data/local/tmp/
adb push imgs /data/local/tmp/
# Run on device
adb shell
cd /data/local/tmp
chmod +x yolo11_demo
export LD_LIBRARY_PATH=/vendor/lib64 or (/vendor/lib)
# Usage: ./yolo11_demo <model_path> <image_dir>
./yolo11_demo yolov11n_int8_A311D2.adla ./imgs
```
**Note:** Replace `yolov11n_int8_A311D2.adla` with your actual model file path.
### Python
**Prerequisites:**
- Python 3.10
- Required packages: `numpy`, `opencv-python`, `amlnnlite`
**Install dependencies:**
```bash
pip install numpy opencv-python amlnnlite-1.0.0-cp310-cp310-linux_aarch64.whl
```
**Run on device:**
```bash
python yolov11.py \
--model-path ./yolov11n_int8_A311D2.adla \
--image-dir ./imgs \
--run-cycles 1 \
--loglevel INFO
```
Argument Descriptions:
| Argument | Description |
| ----------------- | ------------------------------------------------------------ |
| --board-work-path | Work path on board, default is /data/local/tmp |
| --model-path | path to .adla model |
| --image-dir | Directory containing test images |
| --run-cycles | Number of inference cycles, default is 1 |
| --loglevel | Logging level: DEBUG / INFO / WARNING / ERROR, default is WARNING |
The script will automatically process all image files (`.jpg`, `.jpeg`, `.png`, `.bmp`) in the current directory and save results to a `{model_name}_result` folder.
## 5.Results
**Performance Feedback**
By setting the loglevel to INFO, the program provides real-time performance metrics upon completion. The console log will display essential hardware and execution details, including:
- Hardware Information: System and ADLA library versions.
- Model Overview: Basic input/output configurations.
- NPU Metrics: Total inference time (latency) and total DRAM bandwidth consumption.
**Detection Output**
The program will print the detection count. The result image with bounding boxes will be saved to the specified output path (`{model_name}_result`).
You can pull the result folder back to view it:
```bash
adb pull /data/local/tmp/yolov11n_int8_A311D2_result
```
![alt text](result.jpg)
**Profiling Visualization**
After a successful run of the Python demo, a folder named after the model (e.g., `{model_name}`) will be generated in the script directory. This folder contains 5 HTML files that provide a visual and detailed breakdown of per-layer performance:
- `hard_op_chart.html` & `soft_op_chart.html`: Hardware/Software op execution details.
- `dram_rd_chart.html` & `dram_wr_chart.html`: Bandwidth read/write distribution.
- `pie_charts_distribution.html`: Overall resource allocation.
You can pull the result folder back to view it:
```bash
adb pull /data/local/tmp/yolov11n_int8_A311D2
```
Taking hard_op_chart.html as an example (shown below), each layer's ADLA operator name includes parentheses containing the index of the corresponding quantized .tflite layer(s); by default, these indices are suppressed, and operators are labeled generically as "hardware" or "software" without numerical suffixes.
![alt text](Visualization.png)

3
examples/yolov11/cpp/build-android.sh
View file

@ -65,7 +65,8 @@ echo "BUILD_DIR: ${BUILD_DIR}"
mkdir -p ${BUILD_DIR}
cd ${BUILD_DIR}
cmake ../../src \
cmake -Wno-dev ../../src \
-DAMLNN_HOME=${AMLNN_HOME:-} \
-DCMAKE_TOOLCHAIN_FILE=${ANDROID_NDK_PATH}/build/cmake/android.toolchain.cmake \
-DANDROID_ABI=${TARGET_ABI} \
-DANDROID_PLATFORM=android-24 \

79
examples/yolov11/cpp/src/CMakeLists.txt
View file

@ -1,46 +1,35 @@
cmake_minimum_required(VERSION 3.5)
project(yolo11_demo)
set(CMAKE_CXX_STANDARD 17)
# Set NNSDK path
if(NOT DEFINED NNSDK_DIR)
set(NNSDK_DIR "${CMAKE_SOURCE_DIR}/../../../../../amlnn-toolkit/nn_runtime/nnsdk")
endif()
set(NNSDK_ROOT "${NNSDK_DIR}")
message(STATUS "NNSDK_ROOT: ${NNSDK_ROOT}")
include_directories(${NNSDK_ROOT}/include)
include_directories(${CMAKE_SOURCE_DIR}/../../../../common)
# Set dependency path
set(3RDPARTY_DIR "${CMAKE_SOURCE_DIR}/../../../../dependency")
if(CMAKE_SYSTEM_NAME STREQUAL "Android")
if (ANDROID_ABI STREQUAL "arm64-v8a")
link_directories(${NNSDK_ROOT}/android/arm64-v8a)
else()
link_directories(${NNSDK_ROOT}/android/armeabi-v7a)
endif()
# Android needs log
link_libraries(log)
elseif(CMAKE_SYSTEM_NAME STREQUAL "Linux")
link_directories(${NNSDK_ROOT}/linux/yocto/aarch64-poky-linux)
endif()
# Find OpenCV
message(STATUS "OpenCV_DIR: ${OpenCV_DIR}")
find_package(OpenCV REQUIRED)
include_directories(${OpenCV_INCLUDE_DIRS})
add_executable(yolo11_demo
main.cpp
postprocess.cpp
${CMAKE_SOURCE_DIR}/../../../../common/model_loader.cpp
)
target_link_libraries(yolo11_demo
${OpenCV_LIBS}
nnsdk
cmake_minimum_required(VERSION 3.10...3.27)
project(yolo11_demo)
set(CMAKE_CXX_STANDARD 17)
list(APPEND CMAKE_MODULE_PATH "${CMAKE_SOURCE_DIR}/../../../../cmake")
find_package(AMLNN REQUIRED)
include_directories(${AMLNN_INCLUDE_DIR})
link_directories(${AMLNN_LIBRARY_DIR})
include_directories(${CMAKE_SOURCE_DIR}/../../../../common)
# Set dependency path
set(3RDPARTY_DIR "${CMAKE_SOURCE_DIR}/../../../../dependency")
if(CMAKE_SYSTEM_NAME STREQUAL "Android")
# Android needs log
link_libraries(log)
endif()
# Find OpenCV
message(STATUS "OpenCV_DIR: ${OpenCV_DIR}")
find_package(OpenCV REQUIRED)
include_directories(${OpenCV_INCLUDE_DIRS})
add_executable(yolo11_demo
main.cpp
postprocess.cpp
${CMAKE_SOURCE_DIR}/../../../../common/model_loader.cpp
)
target_link_libraries(yolo11_demo
${OpenCV_LIBS}
${AMLNN_LIBRARY}
)

280
examples/yolov8/README.md
View file

@ -1,133 +1,147 @@
# yolov8
## 1.Overview
​ YOLOv8 was released by Ultralytics on January 10, 2023, offering cutting-edge performance in terms of accuracy and speed. Building upon the advancements of previous YOLO versions, YOLOv8 introduced new features and optimizations that make it an ideal choice for various [object detection](https://www.ultralytics.com/blog/a-guide-to-deep-dive-into-object-detection-in-2025) tasks in a wide range of applications.
## 2.Model Download
- **Open Source model**
- **Open Source projects:** https://github.com/ultralytics/ultralytics/tree/v8.2.0
- **Export Model Step:**
- **Install ultralytics**
pip install torch==2.4.1
pip install torchvision==0.19.1
pip install ultralytics==8.2.0
- **Download weights**
wget https://github.com/ultralytics/assets/releases/download/v8.2.0/yolov8m.pt
wget https://github.com/ultralytics/assets/releases/download/v8.2.0/yolov8s.pt
wget https://github.com/ultralytics/assets/releases/download/v8.2.0/yolov8n.pt
- **Export Model**
```
from ultralytics import YOLO
model = YOLO("yolov8m.pt")
model.export(format="onnx", opset=12, simplify=True, dynamic=False, imgsz=640)
```
- **Exported Model**
​ link to amlogic server( **onnx model or quantized tflite**)
## 3. Model Conversion
```
cd model
Usage: ./adla_covnert.sh model_path adla_tookkit_path target_platform
example
./adla_covnert.sh yolov8m.onnx /xxxx/adla-toolkit-binary-3.2.9.3 PRODUCT_PID0XA005
./adla_covnert.sh yolov8s.onnx /xxxx/adla-toolkit-binary-3.2.9.3 PRODUCT_PID0XA005
./adla_covnert.sh yolov8n.onnx /xxxx/adla-toolkit-binary-3.2.9.3 PRODUCT_PID0XA005
```
| Parameter | Discription |
| ----------------- | ------------------------------------------------------------ |
| model_path | onnx model path |
| adla_tookkit_path | path to adla_toolkit |
| target_platform | Specify target platform. for A311D2 : PRODUCT_PID0XA003。for S905X5: PRODUCT_PID0XA005 |
## 4. Demo Run
### CPP
#### 1. Compile
**Prerequisites:**
- Android NDK (r25e recommended)
- `ANDROID_NDK_PATH` environment variable set
**Build:**
```bash
# Build for arm64-v8a
cd examples/yolov8/cpp
./build-android.sh -a arm64-v8a
```
The executable will be generated at `build/android/yolov8_demo` (Note: executable name may vary, verify in build folder).
#### 2. Run
```bash
# Push executable to device
adb push build/android/yolov8_demo /data/local/tmp/
adb push model/yolov8s_int8_A311D2.adla /data/local/tmp/
adb push test_image.jpg /data/local/tmp/
# Run on device
adb shell
cd /data/local/tmp
chmod +x yolov8_demo
export LD_LIBRARY_PATH=/vendor/lib64 or (/vendor/lib)
# Usage: ./yolo_world_demo <model_path> <image_path>
./yolov8_demo yolov8s_int8_A311D2.adla test_image.jpg"
```
**Note:** Replace `yolov8s_int8_A311D2.adla` with your actual model file path.
### Python
**Prerequisites:**
- Python 3.10
- Required packages: `numpy`, `opencv-python`, `amlnnlite`
**Install dependencies:**
```bash
pip install numpy opencv-python amlnnlite-1.0.0-cp310-cp310-linux_aarch64.whl
```
**Run on device:**
```bash
python yolov8.py --model-path ./yolov8s_int8_A311D2.adla
```
The script will automatically process all image files (`.jpg`, `.jpeg`, `.png`, `.bmp`) in the current directory and save results to a `{model_name}_result` folder.
## 5.Results
The program will print the detection count and inference time. The result image with bounding boxes will be saved to the specified output path (`result.jpg` by default).
You can pull the result image back to view it:
```bash
adb pull result.jpg.
```
![alt text](result.jpg)
# yolov8
## 1.Overview
​ YOLOv8 was released by Ultralytics on January 10, 2023, offering cutting-edge performance in terms of accuracy and speed. Building upon the advancements of previous YOLO versions, YOLOv8 introduced new features and optimizations that make it an ideal choice for various [object detection](https://www.ultralytics.com/blog/a-guide-to-deep-dive-into-object-detection-in-2025) tasks in a wide range of applications.
## 2.Model Download
- **Open Source model**
- **Open Source projects:** https://github.com/ultralytics/ultralytics/tree/v8.2.0
- **Export Model Step:**
- **Install ultralytics**
pip install torch==2.4.1
pip install torchvision==0.19.1
pip install ultralytics==8.2.0
- **Download weights**
wget https://github.com/ultralytics/assets/releases/download/v8.2.0/yolov8m.pt
wget https://github.com/ultralytics/assets/releases/download/v8.2.0/yolov8s.pt
wget https://github.com/ultralytics/assets/releases/download/v8.2.0/yolov8n.pt
- **Export Model**
```
from ultralytics import YOLO
model = YOLO("yolov8m.pt")
model.export(format="onnx", opset=12, simplify=True, dynamic=False, imgsz=640)
```
- **Exported Model**
​ link to amlogic server( **onnx model or quantized tflite**)
## 3. Model Conversion
```
cd model
Usage: ./adla_covnert.sh model_path adla_tookkit_path target_platform
example
./adla_covnert.sh yolov8m.onnx /xxxx/adla-toolkit-binary-3.2.9.3 PRODUCT_PID0XA005
./adla_covnert.sh yolov8s.onnx /xxxx/adla-toolkit-binary-3.2.9.3 PRODUCT_PID0XA005
./adla_covnert.sh yolov8n.onnx /xxxx/adla-toolkit-binary-3.2.9.3 PRODUCT_PID0XA005
```
| Parameter | Discription |
| ----------------- | ------------------------------------------------------------ |
| model_path | onnx model path |
| adla_tookkit_path | path to adla_toolkit |
| target_platform | Specify target platform. for A311D2 : PRODUCT_PID0XA003。for S905X5: PRODUCT_PID0XA005 |
## 4. Demo Run
### CPP
#### 1. Compile
#### AMLNN SDK Setup
Resolve the AMLNN nnsdk dependency using one of the following methods:
- **Priority 1 – Environment variable (recommended)**
```bash
export AMLNN_HOME=/path/to/amlnn-toolkit
```
- **Priority 3 – Sibling directory fallback** *(automatic)*
Place `amlnn-toolkit` as a sibling to `amlnn-model-playground`:
```bash
git clone git@github.com:Amlogic-NN/amlnn-toolkit.git ../amlnn-toolkit
```
**Prerequisites:**
- Android NDK (r25e recommended)
- `ANDROID_NDK_PATH` environment variable set
**Build:**
```bash
# Build for arm64-v8a
cd examples/yolov8/cpp
./build-android.sh -a arm64-v8a
```
The executable will be generated at `build/android/yolov8_demo` (Note: executable name may vary, verify in build folder).
#### 2. Run
```bash
# Push executable to device
adb push build/android/yolov8_demo /data/local/tmp/
adb push model/yolov8s_int8_A311D2.adla /data/local/tmp/
adb push test_image.jpg /data/local/tmp/
# Run on device
adb shell
cd /data/local/tmp
chmod +x yolov8_demo
export LD_LIBRARY_PATH=/vendor/lib64 or (/vendor/lib)
# Usage: ./yolo_world_demo <model_path> <image_path>
./yolov8_demo yolov8s_int8_A311D2.adla test_image.jpg"
```
**Note:** Replace `yolov8s_int8_A311D2.adla` with your actual model file path.
### Python
**Prerequisites:**
- Python 3.10
- Required packages: `numpy`, `opencv-python`, `amlnnlite`
**Install dependencies:**
```bash
pip install numpy opencv-python amlnnlite-1.0.0-cp310-cp310-linux_aarch64.whl
```
**Run on device:**
```bash
python yolov8.py --model-path ./yolov8s_int8_A311D2.adla
```
The script will automatically process all image files (`.jpg`, `.jpeg`, `.png`, `.bmp`) in the current directory and save results to a `{model_name}_result` folder.
## 5.Results
The program will print the detection count and inference time. The result image with bounding boxes will be saved to the specified output path (`result.jpg` by default).
You can pull the result image back to view it:
```bash
adb pull result.jpg.
```
![alt text](result.jpg)

4
examples/yolov8/cpp/build-android.sh
View file

@ -65,13 +65,13 @@ echo "BUILD_DIR: ${BUILD_DIR}"
mkdir -p ${BUILD_DIR}
cd ${BUILD_DIR}
cmake ../../src \
cmake -Wno-dev ../../src \
-DAMLNN_HOME=${AMLNN_HOME:-} \
-DCMAKE_TOOLCHAIN_FILE=${ANDROID_NDK_PATH}/build/cmake/android.toolchain.cmake \
-DANDROID_ABI=${TARGET_ABI} \
-DANDROID_PLATFORM=android-24 \
-DCMAKE_BUILD_TYPE=Release \
-DOpenCV_DIR=${ROOT_PWD}/../../../dependency/opencv/opencv-android-sdk-build/sdk/native/jni/abi-${TARGET_ABI} \
-DNNSDK_DIR=${ROOT_PWD}/../../../../amlnn-toolkit/nn_runtime/nnsdk
make -j4

82
examples/yolov8/cpp/src/CMakeLists.txt
View file

@ -1,46 +1,36 @@
cmake_minimum_required(VERSION 3.5)
project(yolo_world_demo)
set(CMAKE_CXX_STANDARD 17)
# Set NNSDK path
if(NOT DEFINED NNSDK_DIR)
set(NNSDK_DIR "${CMAKE_SOURCE_DIR}/../../../../../amlnn-toolkit/nn_runtime/nnsdk")
endif()
set(NNSDK_ROOT "${NNSDK_DIR}")
message(STATUS "NNSDK_ROOT: ${NNSDK_ROOT}")
include_directories(${NNSDK_ROOT}/include)
include_directories(${CMAKE_SOURCE_DIR}/../../../../common)
# Set 3rdparty path
set(3RDPARTY_DIR "${CMAKE_SOURCE_DIR}/../../../../dependency")
if(CMAKE_SYSTEM_NAME STREQUAL "Android")
if (ANDROID_ABI STREQUAL "arm64-v8a")
link_directories(${NNSDK_ROOT}/android/arm64-v8a)
else()
link_directories(${NNSDK_ROOT}/android/armeabi-v7a)
endif()
# Android needs log
link_libraries(log)
elseif(CMAKE_SYSTEM_NAME STREQUAL "Linux")
link_directories(${NNSDK_ROOT}/linux/yocto/aarch64-poky-linux)
endif()
# Find OpenCV
message(STATUS "OpenCV_DIR: ${OpenCV_DIR}")
find_package(OpenCV REQUIRED)
include_directories(${OpenCV_INCLUDE_DIRS})
add_executable(yolov8_demo
main.cpp
postprocess.cpp
postprocess.h
${CMAKE_SOURCE_DIR}/../../../../common/model_loader.cpp
)
target_link_libraries(yolov8_demo
${OpenCV_LIBS}
nnsdk
)
cmake_minimum_required(VERSION 3.10...3.27)
project(yolo_world_demo)
set(CMAKE_CXX_STANDARD 17)
list(APPEND CMAKE_MODULE_PATH "${CMAKE_SOURCE_DIR}/../../../../cmake")
find_package(AMLNN REQUIRED)
include_directories(${AMLNN_INCLUDE_DIR})
link_directories(${AMLNN_LIBRARY_DIR})
include_directories(${CMAKE_SOURCE_DIR}/../../../../common)
# Set 3rdparty path
set(3RDPARTY_DIR "${CMAKE_SOURCE_DIR}/../../../../dependency")
if(CMAKE_SYSTEM_NAME STREQUAL "Android")
# Android needs log
link_libraries(log)
endif()
# Find OpenCV
message(STATUS "OpenCV_DIR: ${OpenCV_DIR}")
find_package(OpenCV REQUIRED)
include_directories(${OpenCV_INCLUDE_DIRS})
add_executable(yolov8_demo
main.cpp
postprocess.cpp
postprocess.h
${CMAKE_SOURCE_DIR}/../../../../common/model_loader.cpp
)
target_link_libraries(yolov8_demo
${OpenCV_LIBS}
${AMLNN_LIBRARY}
)

158
examples/yoloworld/README.md
View file

@ -1,72 +1,86 @@
## Demo Run
### CPP
#### 1. Compile
**Prerequisites:**
- Android NDK (r25e recommended)
- `ANDROID_NDK_PATH` environment variable set
**Build:**
```bash
# Build for arm64-v8a
cd examples/yoloworld/cpp
./build-android.sh -a arm64-v8a
```
The executable will be generated at `build/android/yolo_world_demo` (Note: executable name may vary, verify in build folder).
#### 2. Run
```bash
# Push executable to device
adb push build/android/yolo_world_demo /data/local/tmp/
adb push model/yoloworld_int8_A311D2.adla /data/local/tmp/
adb push test_image.jpg /data/local/tmp/
# Run on device
adb shell
cd /data/local/tmp
chmod +x yolo_world_demo
export LD_LIBRARY_PATH=/vendor/lib64 or (/vendor/lib)
# Usage: ./yolo_world_demo <model_path> <image_path>
./yolo_world_demo yoloworld_int8_A311D2.adla test_image.jpg
```
**Note:** Replace `yoloworld_int8_A311D2.adla` with your actual model file path.
### Python
**Prerequisites:**
- Python 3.10
- Required packages: `numpy`, `opencv-python`, `amlnnlite`
**Install dependencies:**
```bash
pip install numpy opencv-python amlnnlite-1.0.0-cp310-cp310-linux_aarch64.whl
```
**Run on device:**
```bash
# Basic usage (process current directory)
python yoloworld.py --model-path ./yoloworld_int8_A311D2.adla
# Specify image directory
python yoloworld.py --model-path ./yoloworld_int8_A311D2.adla --image-dir ./
```
The script will automatically process all image files (`.jpg`, `.jpeg`, `.png`, `.bmp`) in the specified directory and save results to a `{model_name}_result` folder.
## Results
The program will print the detection count and detected objects for each processed image. The result image with bounding boxes will be saved to the specified output directory.
You can pull the result image back to view it:
```bash
adb pull result.jpg.
```
![alt text](result.jpg)
The program detects objects from predefined classes (handbag, backpack, wallet, watch, necklace, bracelet, earrings, finger ring, sunglass, hat, shoes, belt, makeup palette, lipstick tube, car, truck, bicycle, motorcycle, phone, laptop, camera, wine bottle, stuffed toy) and draws bounding boxes with class labels on the result images.
## Demo Run
### CPP
#### 1. Compile
#### AMLNN SDK Setup
Resolve the AMLNN nnsdk dependency using one of the following methods:
- **Priority 1 – Environment variable (recommended)**
```bash
export AMLNN_HOME=/path/to/amlnn-toolkit
```
- **Priority 3 – Sibling directory fallback** *(automatic)*
Place `amlnn-toolkit` as a sibling to `amlnn-model-playground`:
```bash
git clone git@github.com:Amlogic-NN/amlnn-toolkit.git ../amlnn-toolkit
```
**Prerequisites:**
- Android NDK (r25e recommended)
- `ANDROID_NDK_PATH` environment variable set
**Build:**
```bash
# Build for arm64-v8a
cd examples/yoloworld/cpp
./build-android.sh -a arm64-v8a
```
The executable will be generated at `build/android/yolo_world_demo` (Note: executable name may vary, verify in build folder).
#### 2. Run
```bash
# Push executable to device
adb push build/android/yolo_world_demo /data/local/tmp/
adb push model/yoloworld_int8_A311D2.adla /data/local/tmp/
adb push test_image.jpg /data/local/tmp/
# Run on device
adb shell
cd /data/local/tmp
chmod +x yolo_world_demo
export LD_LIBRARY_PATH=/vendor/lib64 or (/vendor/lib)
# Usage: ./yolo_world_demo <model_path> <image_path>
./yolo_world_demo yoloworld_int8_A311D2.adla test_image.jpg
```
**Note:** Replace `yoloworld_int8_A311D2.adla` with your actual model file path.
### Python
**Prerequisites:**
- Python 3.10
- Required packages: `numpy`, `opencv-python`, `amlnnlite`
**Install dependencies:**
```bash
pip install numpy opencv-python amlnnlite-1.0.0-cp310-cp310-linux_aarch64.whl
```
**Run on device:**
```bash
# Basic usage (process current directory)
python yoloworld.py --model-path ./yoloworld_int8_A311D2.adla
# Specify image directory
python yoloworld.py --model-path ./yoloworld_int8_A311D2.adla --image-dir ./
```
The script will automatically process all image files (`.jpg`, `.jpeg`, `.png`, `.bmp`) in the specified directory and save results to a `{model_name}_result` folder.
## Results
The program will print the detection count and detected objects for each processed image. The result image with bounding boxes will be saved to the specified output directory.
You can pull the result image back to view it:
```bash
adb pull result.jpg.
```
![alt text](result.jpg)
The program detects objects from predefined classes (handbag, backpack, wallet, watch, necklace, bracelet, earrings, finger ring, sunglass, hat, shoes, belt, makeup palette, lipstick tube, car, truck, bicycle, motorcycle, phone, laptop, camera, wine bottle, stuffed toy) and draws bounding boxes with class labels on the result images.

3
examples/yoloworld/cpp/build-android.sh
View file

@ -65,7 +65,8 @@ echo "BUILD_DIR: ${BUILD_DIR}"
mkdir -p ${BUILD_DIR}
cd ${BUILD_DIR}
cmake ../../src \
cmake -Wno-dev ../../src \
-DAMLNN_HOME=${AMLNN_HOME:-} \
-DCMAKE_TOOLCHAIN_FILE=${ANDROID_NDK_PATH}/build/cmake/android.toolchain.cmake \
-DANDROID_ABI=${TARGET_ABI} \
-DANDROID_PLATFORM=android-24 \

80
examples/yoloworld/cpp/src/CMakeLists.txt
View file

@ -1,45 +1,35 @@
cmake_minimum_required(VERSION 3.5)
project(yolo_world_demo)
set(CMAKE_CXX_STANDARD 17)
# Set NNSDK path
if(NOT DEFINED NNSDK_DIR)
set(NNSDK_DIR "${CMAKE_SOURCE_DIR}/../../../../../amlnn-toolkit/nn_runtime/nnsdk")
endif()
set(NNSDK_ROOT "${NNSDK_DIR}")
message(STATUS "NNSDK_ROOT: ${NNSDK_ROOT}")
include_directories(${NNSDK_ROOT}/include)
include_directories(${CMAKE_SOURCE_DIR}/../../../../common)
# Set 3rdparty path
set(3RDPARTY_DIR "${CMAKE_SOURCE_DIR}/../../../../dependency")
if(CMAKE_SYSTEM_NAME STREQUAL "Android")
if (ANDROID_ABI STREQUAL "arm64-v8a")
link_directories(${NNSDK_ROOT}/android/arm64-v8a)
else()
link_directories(${NNSDK_ROOT}/android/armeabi-v7a)
endif()
# Android needs log
link_libraries(log)
elseif(CMAKE_SYSTEM_NAME STREQUAL "Linux")
link_directories(${NNSDK_ROOT}/linux/yocto/aarch64-poky-linux)
endif()
# Find OpenCV
find_package(OpenCV REQUIRED)
include_directories(${OpenCV_INCLUDE_DIRS})
add_executable(yolo_world_demo
main.cpp
postprocess.cpp
postprocess.h
${CMAKE_SOURCE_DIR}/../../../../common/model_loader.cpp
)
target_link_libraries(yolo_world_demo
${OpenCV_LIBS}
nnsdk
)
cmake_minimum_required(VERSION 3.10...3.27)
project(yolo_world_demo)
set(CMAKE_CXX_STANDARD 17)
list(APPEND CMAKE_MODULE_PATH "${CMAKE_SOURCE_DIR}/../../../../cmake")
find_package(AMLNN REQUIRED)
include_directories(${AMLNN_INCLUDE_DIR})
link_directories(${AMLNN_LIBRARY_DIR})
include_directories(${CMAKE_SOURCE_DIR}/../../../../common)
# Set 3rdparty path
set(3RDPARTY_DIR "${CMAKE_SOURCE_DIR}/../../../../dependency")
if(CMAKE_SYSTEM_NAME STREQUAL "Android")
# Android needs log
link_libraries(log)
endif()
# Find OpenCV
find_package(OpenCV REQUIRED)
include_directories(${OpenCV_INCLUDE_DIRS})
add_executable(yolo_world_demo
main.cpp
postprocess.cpp
postprocess.h
${CMAKE_SOURCE_DIR}/../../../../common/model_loader.cpp
)
target_link_libraries(yolo_world_demo
${OpenCV_LIBS}
${AMLNN_LIBRARY}
)

3
examples/yolox/cpp/build-android.sh
View file

@ -65,7 +65,8 @@ echo "BUILD_DIR: ${BUILD_DIR}"
mkdir -p ${BUILD_DIR}
cd ${BUILD_DIR}
cmake ../../src \
cmake -Wno-dev ../../src \
-DAMLNN_HOME=${AMLNN_HOME:-} \
-DCMAKE_TOOLCHAIN_FILE=${ANDROID_NDK_PATH}/build/cmake/android.toolchain.cmake \
-DANDROID_ABI=${TARGET_ABI} \
-DANDROID_PLATFORM=android-24 \

23
examples/yolox/cpp/src/CMakeLists.txt
View file

@ -1,31 +1,21 @@
cmake_minimum_required(VERSION 3.5)
cmake_minimum_required(VERSION 3.10...3.27)
project(yolo11_demo)
set(CMAKE_CXX_STANDARD 17)
# Set NNSDK path
if(NOT DEFINED NNSDK_DIR)
set(NNSDK_DIR "${CMAKE_SOURCE_DIR}/../../../../../amlnn-toolkit/nn_runtime/nnsdk")
endif()
set(NNSDK_ROOT "${NNSDK_DIR}")
message(STATUS "NNSDK_ROOT: ${NNSDK_ROOT}")
list(APPEND CMAKE_MODULE_PATH "${CMAKE_SOURCE_DIR}/../../../../cmake")
find_package(AMLNN REQUIRED)
include_directories(${AMLNN_INCLUDE_DIR})
link_directories(${AMLNN_LIBRARY_DIR})
include_directories(${NNSDK_ROOT}/include)
include_directories(${CMAKE_SOURCE_DIR}/../../../../common)
# Set dependency path
set(3RDPARTY_DIR "${CMAKE_SOURCE_DIR}/../../../../dependency")
if(CMAKE_SYSTEM_NAME STREQUAL "Android")
if (ANDROID_ABI STREQUAL "arm64-v8a")
link_directories(${NNSDK_ROOT}/android/arm64-v8a)
else()
link_directories(${NNSDK_ROOT}/android/armeabi-v7a)
endif()
# Android needs log
link_libraries(log)
elseif(CMAKE_SYSTEM_NAME STREQUAL "Linux")
link_directories(${NNSDK_ROOT}/linux/yocto/aarch64-poky-linux)
endif()
# Find OpenCV
@ -33,7 +23,6 @@ message(STATUS "OpenCV_DIR: ${OpenCV_DIR}")
find_package(OpenCV REQUIRED)
include_directories(${OpenCV_INCLUDE_DIRS})
add_executable(yolox_demo
main.cpp
postprocess.cpp
@ -42,5 +31,5 @@ add_executable(yolox_demo
target_link_libraries(yolox_demo
${OpenCV_LIBS}
nnsdk
${AMLNN_LIBRARY}
)