Error Cannot copy to a TensorFlowLite tensor

[AndryCU]

When i traid to process my model got this error:
java.lang.IllegalArgumentException: Cannot copy to a TensorFlowLite tensor (serving_default_conv2d_input:0) with 90000 bytes from a Java Buffer with 22500 bytes.

Here are the libraries i'm using:
implementation 'org.tensorflow:tensorflow-lite:2.8.0'
implementation 'org.tensorflow:tensorflow-lite-support:0.3.1'
implementation 'org.tensorflow:tensorflow-lite-metadata:0.3.1'

Here you can see my model properties:

I'm traing to config the output of this way, i'm not sure if is of this way..
val p = photoFile.data!!.extras!!.get("data") as Bitmap
val tensorImage = TensorImage.fromBitmap(p)
val model = Modelo.newInstance(applicationContext)1
val imageProcessor: ImageProcessor = ImageProcessor.Builder()
.add(ResizeOp(150, 150, ResizeOp.ResizeMethod.NEAREST_NEIGHBOR))
.add(TransformToGrayscaleOp())
.add(NormalizeOp(0f,255f))
.build()
imageProcessor.process(tensorImage)

[mohantym]

Hi @AndryCU ! Can you let us know the results after changing the resize factor to (300,300)(90000 bytes) instead of (150,150)(22500 bytes) in this line ?

.add(ResizeOp(150, 150, ResizeOp.ResizeMethod.NEAREST_NEIGHBOR))

Attaching relevant thread for reference. R 1 ,2

[AndryCU]

@mohantym it's works but, why change that value if my model was trained with images of 150x150 in greyscale with pixels values between 0 and 1???? Can you explain me??? Also who can i set each pixel value from 0 to1?? I will really apreciet it

[mohantym]

@AndryCU ! The tensor buffer size is determined by datasize (float32: 4bytes) * flat size of the tensor shape (1 * height * width * 1). (4x22500 = 90000 bytes) . Can you allocate a buffer size of 90000 bytes in above method and let us know?

val imgData = ByteBuffer.allocateDirect(4 * 150 * 150 * 1 * 1);

[AndryCU]

@mohantym sorry, can you explain better, i don't know what to do sorry, if you preffer you can text me on telegram @andryssd, of course i will post evething here, but i want more interchange only if you can

[mohantym]

@AndryCU ! Sorry ,I was proposing these changes to have a buffer size of 90000 bytes automatically. Attaching relevant thread for reference. Thanks!

import org.tensorflow.lite.DataType;
import org.tensorflow.lite.support.image.ImageProcessor;
import org.tensorflow.lite.support.image.TensorImage;
import org.tensorflow.lite.support.image.ops.ResizeOp;
import org.tensorflow.lite.support.common.ops.NormalizeOp;


#val p = photoFile.data!!.extras!!.get("data") as Bitmap
#val tensorImage = TensorImage.fromBitmap(p)

ImageProcessor imageProcessor =
    new ImageProcessor.Builder()
       .add(new ResizeOp(150, 150, ResizeOp.ResizeMethod.NEAREST_NEIGHBOR)))
       .add(TransformToGrayscaleOp())
       .add(NormalizeOp(0,255))
       .build();

// Create a TensorImage object. This creates the tensor of the corresponding
// tensor type (Float32 in this case) that the TensorFlow Lite interpreter needs.
TensorImage tensorImage = new TensorImage(DataType.FLOAT32); // setting Datatype float32 to allocate buffer size 90000 automatically

// Analysis code for every frame
// Preprocess the image
tensorImage.load(bitmap);
tensorImage = imageProcessor.process(tensorImage);

[AndryCU]

It's work perfectly. Thanks, i only have to said that you wrote some lines in java, but, android studio detects it and convert to kotlin automatically.

[google-ml-butler]

Are you satisfied with the resolution of your issue?
Yes
No

[YasmineeBa]

@AndryCU hi, i'm getting the same error that you got when i try to run ly model (tflite) in android studio for detecting object,
i got this problem : java.lang.IllegalArgumentException: Cannot copy from a TensorFlowLite tensor (StatefulPartitionedCall:0) with 7616000 bytes to a Java Buffer with 1411200 bytes.

Can you please tell me which file i need to modified to solve this issue please?

[mouathayed]

hey, today I fixed the same issue. In my case, I stored the result of the preprocessing function in a Uint8List object, while my model expects a Float32 object instead. This size difference (4 bytes for float32 vs 1 byte for Uint8List) is causing the problem. I believe you are facing a similar issue because your model is expecting 90000 bytes but receiving only 22500, which is exactly a quarter of what it should be.