torch vision toImmutableImage meet error ,the tensor toArray byte shape length is not the same with the image bytes make ImageParseException

[mullerhai]

Hi,
Now ,we try to use Storch vision to process image for tenso , I know this part dependency with scrimage library. but meet ImageParseException: Image parsing failed for unknown image format ，from image to Tensor then convert to Image

I know the demo image have 3 channels colorful image , so get the pixel matrix / tensor shape is [3, 1199, 907],so get the byte length is 3* 1199 * 907 = 3262479 but use the
val image = ImmutableImage.loader().fromFile(file)

val bytes = image.forWriter(JpegWriter.Default).bytes()
val outImage = ImmutableImage.loader().fromBytes(bytes)
outImage.output(JpegWriter.Default, new File("src/main/resources/flipx.jpeg"))
, get the bytes length is 147369

object functional {
  def toTensor(pic: ImmutableImage): Tensor[Float32] =
    val bytes = pic.rgb.flatten
    // transpose NxHxWxC to NxCxHxW because pytorch expects channels first
    Tensor(ArraySeq.unsafeWrapArray(bytes))
      .reshape(pic.height, pic.width, 3)
      .permute(2, 0, 1)
      .to(dtype = float32) / 255

  def toImmutableImage[D <: FloatNN](pic: Tensor[D]): ImmutableImage =
    var _pic = pic
    if !Seq(2, 3).contains(pic.dim) then
      throw new IllegalArgumentException(
        s"pic should be 2/3 dimensional. Got ${pic.dim} dimensions."
      )
    else if pic.dim == 2 then
      // if 2D image, add channel dimension (CHW)
      _pic = pic.unsqueeze(0)
    // check number of channels
    println(pic.shape)
    if pic.shape(-3) > 4 then
      throw new IllegalArgumentException(
        s"pic should not have > 4 channels. Got ${pic.shape(-3)} channels."
      )
    val intImage = (_pic.permute(1, 2, 0) * 255).to(dtype = int8)
    val bytes = intImage.toArray
    ImmutableImage.loader().fromBytes(bytes)
}

  import java.io.File

extension (pic: ImmutableImage)
    def toTensors() = toTensor(pic)

  def grayscale(pic: ImmutableImage, p: GrayscaleMethod = GrayscaleMethod.AVERAGE): Tensor[Float32] =
    pic.toGrayscale(p).toTensors()

  val file = new File("src/main/resources/a1.jpg")

  val image = ImmutableImage.loader().fromFile(file)

  val tensor2 = functional.grayscale(image)

  println(tensor9)
  val outImage =functional.toImmutableImage(tensor9)
  outImage.output(JpegWriter.Default, new File("src/main/resources/flipx.jpeg"))

error console log

tensor dtype=float32, shape=[3, 1199, 907], device=CPU 
[[[-0.6965, -0.0801, 0.5022, ..., -0.3883, -0.4054, -0.4054],
  [-0.3883, 0.6221, 1.3413, ..., -0.4054, -0.4054, -0.4054],
  [0.0227, 0.0569, -0.1999, ..., -0.2856, -0.2856, -0.3027],
  ...,
  [0.9474, 0.9303, 0.9132, ..., 1.2557, 1.2214, 1.1872],
  [1.0502, 1.0331, 0.9988, ..., 1.1700, 1.1015, 1.0331],
  [1.1015, 1.0844, 1.0502, ..., 1.5125, 1.4440, 1.3584]],

 [[-0.5826, 0.0476, 0.6429, ..., -0.2675, -0.2850, -0.2850],
  [-0.2675, 0.7654, 1.5007, ..., -0.2850, -0.2850, -0.2850],
  [0.1527, 0.1877, -0.0749, ..., -0.1625, -0.1625, -0.1800],
  ...,
  [1.0980, 1.0805, 1.0630, ..., 1.4132, 1.3782, 1.3431],
  [1.2031, 1.1856, 1.1506, ..., 1.3256, 1.2556, 1.1856],
  [1.2556, 1.2381, 1.2031, ..., 1.6758, 1.6057, 1.5182]],

 [[-0.3578, 0.2696, 0.8622, ..., -0.0441, -0.0615, -0.0615],
  [-0.0441, 0.9842, 1.7163, ..., -0.0615, -0.0615, -0.0615],
  [0.3742, 0.4091, 0.1476, ..., 0.0605, 0.0605, 0.0431],
  ...,
  [1.3154, 1.2980, 1.2805, ..., 1.6291, 1.5942, 1.5594],
  [1.4200, 1.4025, 1.3677, ..., 1.5420, 1.4722, 1.4025],
  [1.4722, 1.4548, 1.4200, ..., 1.8905, 1.8208, 1.7337]]]
ArraySeq(3, 1199, 907)
Exception in thread "main" com.sksamuel.scrimage.ImageParseException: Image parsing failed for unknown image format. Tried the following ImageReader implementations:
com.sksamuel.scrimage.nio.ImageIOReader (delegates to the JDK javax.imageio readers) failed due to No javax.imageio.ImageReader supported this image format
com.sksamuel.scrimage.nio.PngReader (delegates to ar.com.hjg.pngj) failed
com.sksamuel.scrimage.webp.WebpImageReader@37864b77 failed due to java.io.IOException: [Decoding of C:\Users\hai71\AppData\Local\Temp\input6936903937282349564webp failed., Status: 3(BITSTREAM_ERROR)]
	at com.sksamuel.scrimage.nio.ImageReaders.read(ImageReaders.java:64)
	at com.sksamuel.scrimage.nio.ImageReaders.read(ImageReaders.java:37)
	at com.sksamuel.scrimage.nio.ImmutableImageLoader.load(ImmutableImageLoader.java:150)
	at com.sksamuel.scrimage.nio.ImmutableImageLoader.fromBytes(ImmutableImageLoader.java:103)
	at torchvision.transforms.functional$.toImmutableImage(functional.scala:650)
	at main$package$.main(main.scala:27)
	at main.main(main.scala:5)