JpegEncoder

class torchcodec.encoders.JpegEncoder(img: Tensor)

Encoder for JPEG images.

For a tutorial, see: Encoding images.

Example

from torchcodec.encoders import JpegEncoder

JpegEncoder(img).to_file("image.jpg")
# or encode to a file-like object or to a tensor, see methods below.

Parameters:

img (torch.Tensor) – The image to encode, a 3-dimensional uint8 tensor in CHW layout with 1 (grayscale) or 3 (RGB) channels. If on a CUDA device, encoding is performed on the GPU with nvJPEG, and only 3-channel RGB is supported.

Examples using JpegEncoder:

Encoding images

Encoding images

Migrating from TorchVision to TorchCodec

Migrating from TorchVision to TorchCodec

to_file(dest: str | Path, *, quality: int = 75) → None

Encode the image into a JPEG file.

Parameters:

• dest (str or pathlib.Path) – The path to the output file, e.g. image.jpg.

• quality (int, optional) – Quality of the output, between 1 and 100. Higher means better quality and larger file size. Default: 75.

to_file_like(dest: RawIOBase | BufferedIOBase, *, quality: int = 75) → None

Encode the image into a file-like object.

Parameters:

• dest – A writable file-like object supporting write and seek, such as io.BytesIO() or an open file in binary write mode.

• quality (int, optional) – Quality of the output, between 1 and 100. Higher means better quality and larger file size. Default: 75.

to_tensor(*, quality: int = 75) → Tensor

Encode the image into raw bytes, as a 1D uint8 tensor.

The returned tensor is on the same device as the input image (a CUDA input yields a CUDA tensor).

Parameters:

quality (int, optional) – Quality of the output, between 1 and 100. Higher means better quality and larger file size. Default: 75.

Returns:

The encoded bytes, a 1D uint8 tensor on the same device as the input image.

Return type:

torch.Tensor