SNPE 教程 - 基本用法整理 (0)

1. create input

由于接收的是tensor，所以需要将图转化为tensor作为输入：

//将单通道转化为float数组
private float[] convertBitmapToByteBuffer(Bitmap bitmap) {
        private int[] intValues = new int[IMAGE_WIDTH * IMAGE_HEIGHT];
        int lengthValues = IMAGE_WIDTH * IMAGE_HEIGHT * 3;
        float[] floatValues = new float[lengthValues];
        FloatBuffer floatBuffer = FloatBuffer.wrap(floatValues, 0, lengthValues);
        floatBuffer.rewind();
        bitmap.getPixels(intValues, 0, bitmap.getWidth(), 0, 0, bitmap.getWidth(), bitmap.getHeight());
        // Convert the [grayScaled] image to floating point.
        int pixel = 0;
        for (int i = 0; i < getImageSizeX(); ++i) {
            for (int j = 0; j < getImageSizeY(); ++j) {
                final int val = intValues[pixel++];
                floatBuffer.put(pixelValue & 0xFF);
                floatBuffer.put((pixelValue >> 8) & 0xFF);
                floatBuffer.put((pixelValue >> 16) & 0xFF);
            }
        }
        return floatBuffer.toArray();
    }
//RGB三通直接将BGR的数据传入生成 IMAGE_WIDTH * IMAGE_HEIGHT * 3的数组即可。

将图像数据转化为TensorFloat:

 //传入输入层的Dimension
 final FloatTensor tensor = mNeuralNetwork.createFloatTensor(1, 192, 192, 3);
 //将以上图片数据写入tensor
 tensor.write(input, 0, input.length);
 //创建inputs
final Map<String, FloatTensor> inputs = new HashMap<>();
//传入Model的input Layer的名称（注意tensorFlow的模型需要加上输出的位置索引 DATA_INPUT_LAYER_NAME = "input:0"）
inputs.put(DATA_INPUT_LAYER_NAME, tensor);

2. create output

指定模型的output Layer的名称:索引

//forward
final Map<String, FloatTensor> outputs = mNeuralNetwork.execute(inputs);

3. parse results

//该模型的输出层Dimension为 1*96 * 96 * 14
float resultOut[] = new float[96 * 96 * 14];
for (Map.Entry<String, FloatTensor> output : outputs.entrySet()) {
            //OUTPUT_LAYER_NAME = "out_5_stage:0"
            if (output.getKey().equals(OUTPUT_LAYER_NAME)) {
                FloatTensor outputLayer = output.getValue();
                outputLayer.read(resultOut, 0, 96 * 96 * 14, 0 );
            }
        }

此时结果就都在resultOut之中。