1 Introduction

This is an Android demo app to show how to integrate the output YoloX models by XGen into an android app.

2 Integration of XGen output into an app

Readers can see app/src/main/cpp/inference_api.cc to see how the output of XGen is used in an app for AI. This part gives the explanation.

2.1 Import XGen SDK

Please refer to XGen Document for XGen generated files.

Find *.data, *.pb files from *compiled_file/android/model under XGen workplace, and put them into app/src/main/assets in this project. In this example, they are renamed to yolox_80_640x640.data and yolox_80_640x640.pb.

Modify app/src/main/java/com/cocopie/xgen/yolo/example/Yolox.kt, change the variable XGEN_YOLOX_MODEL_NAME value to the file name of the above *.data *.pb.

Put your label file intoapp/src/main/assets and change the LABEL_YOLOX_FILE_NAME value accordingly.

Find libxgen.so from *compiled_file/android/lib under XGen workplace, and put it into app/src/main/jniLibs/arm64-v8a in this project.

If you want to compare the speed of the original onnx file with the XGen outputs, you can put the onnx file into the app/src/main/assets and toggle it by clicking on the 'USING XGEN' or 'USING ONNX' tab in the top right corner on the app.

You can modify the universalThreshold variable in the app/src/main/java/com/cocopie/xgen/yolo/example/Yolox.kt file to adjust the detection threshold

2.2 Initialize XGen

Call the XGenInitWithFiles method to initialize XGen.

2.3 Run XGen

First call the XGenCopyBufferToTensor method to pass the preprocessed data into XGen runtime.

Then call the XGenRun method to let XGen runtime call the AI model to conduct an inference.

And finally call the XGenCopyTensorToBuffer method to copy the result into a buffer.

    jobjectArray RunInference(JNIEnv *env, const jobjectArray input_data) {
        size_t input_tensor_num = XGenGetNumInputTensors(h);
        for (int i = 0; i < input_tensor_num; ++i) {
            auto array = (jfloatArray) env->GetObjectArrayElement(input_data, i);
            jfloat *array_data = env->GetFloatArrayElements(array, JNI_FALSE);
            XGenTensor *input_tensor = XGenGetInputTensor(h, i);
            size_t input_size_in_bytes = XGenGetTensorSizeInBytes(input_tensor);
            size_t input_size = input_size_in_bytes / sizeof(float);
            LOGI("tensor_index:%d input_size:%zu", i, input_size);
            auto buffer = std::shared_ptr<float>(new float[input_size], std::default_delete<float[]>());
            for (int j = 0; j < input_size; ++j) {
                buffer.get()[j] = array_data[j];
            }
            XGenCopyBufferToTensor(input_tensor, buffer.get(), input_size_in_bytes);
        }

        if (XGenRun(h) != XGenOk) {
            LOGI("FATAL ERROR: XGen inference failed.");
            return nullptr;
        }

        size_t output_tensor_num = XGenGetNumOutputTensors(h);
        jclass floatArrayCls = env->FindClass("[F");
        jobjectArray jReturnData = env->NewObjectArray(output_tensor_num, floatArrayCls, nullptr);
        for (int i = 0; i < output_tensor_num; ++i) {
            XGenTensor *output_tensor = XGenGetOutputTensor(h, i);
            size_t output_size_in_bytes = XGenGetTensorSizeInBytes(output_tensor);
            size_t output_size = output_size_in_bytes / sizeof(float);
            auto output_data = std::shared_ptr<float>(new float[output_size], std::default_delete<float[]>());
            XGenCopyTensorToBuffer(output_tensor, output_data.get(), output_size_in_bytes);
            jfloatArray jOutputData = env->NewFloatArray(output_size);
            env->SetFloatArrayRegion(jOutputData, 0, (jsize) output_size, output_data.get());
            env->SetObjectArrayElement(jReturnData, i, jOutputData);
        }
        return jReturnData;
    }

2.4 Screenshot

2.5 Performance comparison

Faster and smaller model files with almost identical accuracy.

Model	Terminal Latency (ms)	Demo FPS (Snapdragon 888)	Size (MB)
YOLOX(ONNX)	587	1.7	34.2
YOLOX(XGen)	94	10.6	17.1

3 Copyright

© 2022 CoCoPIE Inc. All Rights Reserved.

CoCoPIE Inc., its logo and certain names, product and service names reference herein may be registered trademarks, trademarks, trade names or service marks of CoCoPIE Inc. in certain jurisdictions.

The material contained herein is proprietary, privileged and confidential and owned by CoCoPIE Inc. or its third-party licensors. The information herein is provided only to be person or entity to which it is addressed, for its own use and evaluation; therefore, no disclosure of the content of this manual will be made to any third parities without specific written permission from CoCoPIE Inc.. The content herein is subject to change without further notice. Limitation of Liability – CoCoPIE Inc. shall not be liable.

All other trademarks are the property of their respective owners. Other company and brand products and service names are trademarks or registered trademarks of their respective holders.

Limitation of Liability

CoCoPIE Inc. shall not be liable.