mstankie
commented
4 years ago Hi! It's a bit late, but I have just faced the same problem trying to replicate the instruction.
I am not a Java programmer, but I have found that the size of the array allocated for imgData is reduced, but in convertBitmapToByteBuffer() it is still packing floats into the array.
I have changed the function as shown below and it works now. I am not yet sure whether just replacing the putFloat to put (the latter puts bytes) and just trimming floats to bytes in its argument is correct, but solves the main problem.
private void convertBitmapToByteBuffer(Bitmap bitmap) {
if (imgData == null) {
return;
}
imgData.rewind();
bitmap.getPixels(intValues, 0, bitmap.getWidth(), 0, 0, bitmap.getWidth(), bitmap.getHeight());
// Convert the image to floating point.
int pixel = 0;
long startTime = SystemClock.uptimeMillis();
for (int i = 0; i < DIM_IMG_SIZE_X; ++i) {
for (int j = 0; j < DIM_IMG_SIZE_Y; ++j) {
final int val = intValues[pixel++];
imgData.put((byte)((val >> 16) & 0xFF));
imgData.put((byte)((val >> 8) & 0xFF));
imgData.put((byte)((val) & 0xFF));
}
}
long endTime = SystemClock.uptimeMillis();
Log.d(TAG, "Timecost to put values into ByteBuffer: " + Long.toString(endTime - startTime));
}I have a feeling there is a way of just copying the buffers (or even reusing the Bitmap data array) without looping through x and y.
It is really frustrating to see that the official documentation is not working. I followed the step by step to deploy a TensorFlow Lite model created using AutoML Vision Edge. As usual it works for the provided graph.lite and labels.txt files. However it fails when a custom object detector model exported from AutoML Vision Edge is used. Necessary recommended changes were made. Here is the metadata about the model that was exported from AutoML:
Quantization is there, so the change from INT to Float should be carried out as app uses a float model and I have done those changes as instructed. The gradle were synced and app seems to be built as well. But it crashes momentarily with the following error:
I am not an Android developer, so I am having hard time what causes the crash. I think though that the input image by the camera is not the right input for the model.