Skip to main content

AI Processor and SDK Testing Methods

TFlite test​

Click here to download tflite_test.zip.

Decompress tflite_test.zip and run the following commands:

adb root
adb shell mkdir -p /data/profiling
adb push ./tflite_test/Labels /data
adb push ./tflite_test/Models /data

Use case 1​

adb shell
label_image -l /data/Labels/labels.txt -i /data/Labels/grace_hopper.bmp -m /data/Models/mobilenet_v1_1.0_224.tflite -c 10 -j 1 -p 1

The following output indicates that the execution is successful:

Use case 2​

CPU and XNNPACK (open-source neural network inference library)​

adb shell
benchmark_model --graph=/data/Models/mobilenet_v1_1.0_224.tflite --enable_op_profiling=true --use_gpu=false --num_threads=1 --num_runs=100

The following output indicates that the execution is successful:

CPU only​

adb shell
benchmark_model --graph=/data/Models/mobilenet_v1_1.0_224.tflite --enable_op_profiling=true --use_gpu=false  --use_xnnpack=false --num_threads=1 --num_runs=100

The following output indicates that the execution is successful:

QNN test​

Visit https://github.com/rubikpi-ai/demo to download qnn_test.zip.

Decompress qnn_test.zip and run the script 6490_PI_QNN.sh:

chmod -R 777 ./6490_PI_QNN.sh
./6490_PI_QNN.sh

The following output indicates that the execution is successful:

SNPE test​

Visit https://github.com/rubikpi-ai/demo to download snpe_test.zip.

Decompress snpe_test.zip and run the following command:

adb push ./snpe_test/yolo* /opt

Using CPU​

adb shell
snpe-throughput-net-run --container /opt/yolonas.dlc  --duration 100 --perf_profile "high_performance" --userbuffer_tf8 --userbuffer_float_output --use_cpu

The following output indicates that the execution is successful:

Using DSP​

adb shell
snpe-throughput-net-run --container /opt/yolov5s_v2.5_quantized_cache.dlc  --duration 100 --perf_profile burst --userbuffer_tf8 --userbuffer_float_output --use_dsp

The following output indicates that the execution is successful: