How to extract sample data from pb files?

Ask a Question

Question

Hello, I tried MNIST Handwritten digit recognition in Java( and Scala).

I guessed test_dataset_0/input_0.pb contains sample data of type float array, but I didn't know/couldn't found how to extract that data from Java program. Thus, I attempted to decode input_0.pb using protoc.

After decoding raw input_0.pb file using protoc, I found there are key value pairs, one of which has 9 as key and quite long string as value, which seems to be encoded sample data of float array.

cat input_0.pb | protoc --decode_raw
1: 1
1: 1
1: 28
1: 28
2: 1
9: "\000\000\000\000\000\000\000\000\000\000\000\00........

Then, I found UnknownFieldSet in com.google.protobuf package.

I managed to extract sample data from pb file in the following way.

  val result = ArrayBuilder.ofFloat()
  val f = UnknownFieldSet
    .parseFrom(
      Files.readAllBytes(Path.of("./mnist/test_data_set_0/input_0.pb"))
    )
    .getField(9)
  val buf = f.getLengthDelimitedList().get(0).asReadOnlyByteBuffer()
  val cis = CodedInputStream.newInstance(buf)
  while (!cis.isAtEnd) {
    result.addOne(cis.readFloat())
  }
result.result().toList
// List(0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 3.0, 0.0, 0.0, 4.0, 2.0, 0.0, 11.0, 0.0, 0.0, 14.0, 1.0, 0.0, 19.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 12.0, 0.0, 0.0, 7.0, 0.0, 1.0, 10.0, 0.0, 2.0, 2.0, 16.0, 0.0, 3.0, 3.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 7.0, 8.0, 0.0, 8.0, 0.0, 0.0, 8.0, 0.0, 0.0, 19.0, 0.0, 0.0, 1.0, 21.0, 0.0, 4.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 11.0, 0.0, 0.0, 10.0, 3.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 13.0, 0.0, 15.0, 10.0, 26.0, 34.0, 17.0, 77.0, 181.0, 178.0, 35.0, 4.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 150.0, 254.0, 250.0, 251.0, 243

result.result().toList.length 
// => 784

Is this right way to extract sample data from pb file or is there better way to do this?

Further information

Relevant Area (e.g. model usage, backend, best practices, pre-/post- processing, converters):
model usage, best practives

Is this issue related to a specific model?

Model name (e.g. mnist): MNIST Handwritten digit recognition
Model opset (e.g. 7):

Notes

Any additional information, code snippets.

Start with the linked MNIST model page and its test_data_set_0/input_0.pb file. Review the reported Java and Scala UnknownFieldSet/protoc approach, then document whether it is appropriate or identify the supported extraction method; done means a clear model-usage answer for this sample data.