Struct TensorImage 

pub struct TensorImage { /* private fields */ }

An image represented as a tensor with associated format information.

Implementations

impl TensorImage

pub fn new( width: usize, height: usize, fourcc: FourCharCode, memory: Option<TensorMemory>, ) -> Result<Self>

Creates a new TensorImage with the specified width, height, format, and memory type.

Examples

use edgefirst_image::{RGB, TensorImage};
use edgefirst_tensor::TensorMemory;
let img = TensorImage::new(640, 480, RGB, Some(TensorMemory::Mem))?;
assert_eq!(img.width(), 640);
assert_eq!(img.height(), 480);
assert_eq!(img.fourcc(), RGB);
assert!(!img.is_planar());

pub fn from_tensor(tensor: Tensor<u8>, fourcc: FourCharCode) -> Result<Self>

Creates a new TensorImage from an existing tensor and specified format.

The required tensor shape depends on the pixel format:

Format	Shape	Description
RGB	[H, W, 3]	3-channel interleaved
RGBA	[H, W, 4]	4-channel interleaved
GREY	[H, W, 1]	Single-channel grayscale
YUYV	[H, W, 2]	YUV 4:2:2 interleaved
PLANAR_RGB	[3, H, W]	Channels-first (3 planes)
PLANAR_RGBA	[4, H, W]	Channels-first (4 planes)
RGB_INT8	[H, W, 3]	Packed RGB, int8 via XOR 0x80
PLANAR_RGB_INT8	[3, H, W]	Planar RGB, int8 via XOR 0x80
NV12	[H*3/2, W]	Semi-planar YUV 4:2:0 (2D)
NV16	[H*2, W]	Semi-planar YUV 4:2:2 (2D)

Most formats use a 3D tensor where the channel dimension matches the format’s channel count. The semi-planar formats NV12 and NV16 are special: the Y and UV planes have different heights, so the data cannot be described as [H, W, C]. Instead the contiguous memory is represented as a 2D tensor whose first dimension encodes the total byte height (Y rows + UV rows).

Examples

RGB (3D interleaved):

use edgefirst_image::{RGB, TensorImage};
use edgefirst_tensor::Tensor;
let tensor = Tensor::new(&[720, 1280, 3], None, None)?;
let img = TensorImage::from_tensor(tensor, RGB)?;
assert_eq!(img.width(), 1280);
assert_eq!(img.height(), 720);
assert_eq!(img.fourcc(), RGB);

GREY (3D with 1 channel):

use edgefirst_image::{GREY, TensorImage};
use edgefirst_tensor::Tensor;
let tensor = Tensor::new(&[480, 640, 1], None, None)?;
let img = TensorImage::from_tensor(tensor, GREY)?;
assert_eq!(img.width(), 640);
assert_eq!(img.height(), 480);

NV12 (2D semi-planar, height*3/2 rows):

use edgefirst_image::{NV12, TensorImage};
use edgefirst_tensor::Tensor;
// 1080p NV12: 1080 Y rows + 540 UV rows = 1620 total rows
let tensor = Tensor::new(&[1620, 1920], None, None)?;
let img = TensorImage::from_tensor(tensor, NV12)?;
assert_eq!(img.width(), 1920);
assert_eq!(img.height(), 1080);

pub fn load( image: &[u8], format: Option<FourCharCode>, memory: Option<TensorMemory>, ) -> Result<Self>

Loads an image from the given byte slice, attempting to decode it as JPEG or PNG format. Exif orientation is supported. The default format is RGB.

Examples

use edgefirst_image::{RGBA, TensorImage};
use edgefirst_tensor::TensorMemory;
let jpeg_bytes = include_bytes!("../../../testdata/zidane.png");
let img = TensorImage::load(jpeg_bytes, Some(RGBA), Some(TensorMemory::Mem))?;
assert_eq!(img.width(), 1280);
assert_eq!(img.height(), 720);
assert_eq!(img.fourcc(), RGBA);

pub fn load_jpeg( image: &[u8], format: Option<FourCharCode>, memory: Option<TensorMemory>, ) -> Result<Self>

Loads a JPEG image from the given byte slice. Supports EXIF orientation. The default format is RGB.

Examples

use edgefirst_image::{RGB, TensorImage};
use edgefirst_tensor::TensorMemory;
let jpeg_bytes = include_bytes!("../../../testdata/zidane.jpg");
let img = TensorImage::load_jpeg(jpeg_bytes, Some(RGB), Some(TensorMemory::Mem))?;
assert_eq!(img.width(), 1280);
assert_eq!(img.height(), 720);
assert_eq!(img.fourcc(), RGB);

pub fn load_png( image: &[u8], format: Option<FourCharCode>, memory: Option<TensorMemory>, ) -> Result<Self>

Loads a PNG image from the given byte slice. Supports EXIF orientation. The default format is RGB.

Examples

use edgefirst_image::{RGB, TensorImage};
use edgefirst_tensor::TensorMemory;
let png_bytes = include_bytes!("../../../testdata/zidane.png");
let img = TensorImage::load_png(png_bytes, Some(RGB), Some(TensorMemory::Mem))?;
assert_eq!(img.width(), 1280);
assert_eq!(img.height(), 720);
assert_eq!(img.fourcc(), RGB);

pub fn save_jpeg(&self, path: &str, quality: u8) -> Result<()>

Saves the image as a JPEG file at the specified path with the given quality. Only RGB and RGBA formats are supported.

Examples

use edgefirst_image::{RGB, TensorImage};
use edgefirst_tensor::Tensor;
let tensor = Tensor::new(&[720, 1280, 3], None, None)?;
let img = TensorImage::from_tensor(tensor, RGB)?;
let save_path = "/tmp/output.jpg";
img.save_jpeg(save_path, 90)?;

pub fn tensor(&self) -> &Tensor<u8>

Returns a reference to the underlying tensor.

Examples

use edgefirst_image::{RGB, TensorImage};
use edgefirst_tensor::{Tensor, TensorTrait};
let tensor = Tensor::new(&[720, 1280, 3], None, Some("Tensor"))?;
let img = TensorImage::from_tensor(tensor, RGB)?;
let underlying_tensor = img.tensor();
assert_eq!(underlying_tensor.name(), "Tensor");

pub fn fourcc(&self) -> FourCharCode

Returns the FourCC code representing the image format.

Examples

use edgefirst_image::{RGB, TensorImage};
use edgefirst_tensor::{Tensor, TensorTrait};
let tensor = Tensor::new(&[720, 1280, 3], None, Some("Tensor"))?;
let img = TensorImage::from_tensor(tensor, RGB)?;
assert_eq!(img.fourcc(), RGB);

pub fn is_planar(&self) -> bool

Examples

use edgefirst_image::{RGB, TensorImage};
use edgefirst_tensor::{Tensor, TensorTrait};
let tensor = Tensor::new(&[720, 1280, 3], None, Some("Tensor"))?;
let img = TensorImage::from_tensor(tensor, RGB)?;
assert!(!img.is_planar());

pub fn width(&self) -> usize

Examples

use edgefirst_image::{RGB, TensorImage};
use edgefirst_tensor::{Tensor, TensorTrait};
let tensor = Tensor::new(&[720, 1280, 3], None, Some("Tensor"))?;
let img = TensorImage::from_tensor(tensor, RGB)?;
assert_eq!(img.width(), 1280);

pub fn height(&self) -> usize

Examples

use edgefirst_image::{RGB, TensorImage};
use edgefirst_tensor::{Tensor, TensorTrait};
let tensor = Tensor::new(&[720, 1280, 3], None, Some("Tensor"))?;
let img = TensorImage::from_tensor(tensor, RGB)?;
assert_eq!(img.height(), 720);

pub fn channels(&self) -> usize

Examples

use edgefirst_image::{RGB, TensorImage};
use edgefirst_tensor::{Tensor, TensorTrait};
let tensor = Tensor::new(&[720, 1280, 3], None, Some("Tensor"))?;
let img = TensorImage::from_tensor(tensor, RGB)?;
assert_eq!(img.channels(), 3);

pub fn row_stride(&self) -> usize

Examples

use edgefirst_image::{RGB, TensorImage};
use edgefirst_tensor::{Tensor, TensorTrait};
let tensor = Tensor::new(&[720, 1280, 3], None, Some("Tensor"))?;
let img = TensorImage::from_tensor(tensor, RGB)?;
assert_eq!(img.row_stride(), 1280*3);

pub fn buffer_identity(&self) -> &BufferIdentity

Returns the buffer identity of the underlying tensor.

Trait Implementations

impl Debug for TensorImage

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl TensorImageDst for TensorImage

fn tensor(&self) -> &Tensor<u8>

Returns a reference to the underlying tensor.

fn tensor_mut(&mut self) -> &mut Tensor<u8>

Returns a mutable reference to the underlying tensor.

fn fourcc(&self) -> FourCharCode

Returns the FourCC code representing the image format.

fn is_planar(&self) -> bool

Returns whether the image is in planar format.

fn width(&self) -> usize

Returns the width of the image in pixels.

fn height(&self) -> usize

Returns the height of the image in pixels.

fn channels(&self) -> usize

Returns the number of channels in the image.

fn row_stride(&self) -> usize

Returns the row stride in bytes.

fn buffer_identity(&self) -> &BufferIdentity

Returns the buffer identity of the underlying tensor.

impl TryFrom<&TensorImage> for G2DSurface

type Error = Error

The type returned in the event of a conversion error.

fn try_from(img: &TensorImage) -> Result<Self, Self::Error>

Performs the conversion.

impl TryFrom<&mut TensorImage> for G2DSurface

type Error = Error

The type returned in the event of a conversion error.

fn try_from(img: &mut TensorImage) -> Result<Self, Self::Error>

Performs the conversion.