Grayscale in tensorflow lite

Question

Iam trying to implement tensorflow lite models in android device with grayscale input, but most of the documentation that i found is using rgb input. Is there any example of how to using tflite file in android with grayscale image or it is still not supported ?

When you have the gray scale, you only have 1 channel of an image. The rest is analogical with the rgb image. — Benjamin
i already tried that but there is always an error with BufferOverflowException — Christian Jeremi
Then the problem is probably with the Buffer Overflow, not with the grayscale image. — Benjamin

Hasaan Ahmed Hasaan Ahmed · Accepted Answer · 2020-07-22T01:06:17

Here's how I am converting my RGB bitmap to a corresponding grayscale byte buffer directly:

private ByteBuffer getByteBuffer(Bitmap bitmap){
    int width = bitmap.getWidth();
    int height = bitmap.getHeight();
    ByteBuffer mImgData = ByteBuffer
            .allocateDirect(4 * width * height);
    mImgData.order(ByteOrder.nativeOrder());
    int[] pixels = new int[width*height];
    bitmap.getPixels(pixels, 0, width, 0, 0, width, height);
    for (int pixel : pixels) {
        mImgData.putFloat((float) Color.red(pixel));
    }
    return mImgData;
}

If you need normalized values [0, 1] just divide by 255.0:

float value = (float) Color.red(pixel)/255.0f;
mImgData.putFloat(value);

You can then use this in your interpreter as:

ByteBuffer input = getByteBuffer(bitmap);
tflite.run(input, outputValue);

P.S. I was also getting BufferOverflowException because I was decoding 28x28 image from drawables which was being resized to 56x56 due to dpi conversions. I put it in drawable-nodpi and then it worked correctly. Make sure your image is the right size.

Grayscale in tensorflow lite

1 Answers