Skip to content

unienv_data.transformations.image_compress

Classes:

Name Description
ImageCompressTransformation
ImageDecompressTransformation

Functions:

Name Description
get_jpeg_compression_ratio

Attributes:

Name Type Description
JPEG_QUALITY_COMPRESSION_MAP
CONSERVATIVE_COMPRESSION_RATIOS
CONSERVATIVE_HEADER_BYTES

JPEG_QUALITY_COMPRESSION_MAP module-attribute

JPEG_QUALITY_COMPRESSION_MAP = {
    "quality": np.array([55, 60, 65, 70, 75, 80, 85, 90, 95, 100], dtype=int),
    "compression_ratio": np.array(
        [43.27, 36.9, 34.24, 31.5, 26.0, 25.06, 19.08, 14.3, 9.88, 5.27], dtype=float
    ),
    "conservative_ratio": 0.6,
}

CONSERVATIVE_COMPRESSION_RATIOS module-attribute

CONSERVATIVE_COMPRESSION_RATIOS = {'JPEG': get_jpeg_compression_ratio}

CONSERVATIVE_HEADER_BYTES module-attribute

CONSERVATIVE_HEADER_BYTES = {'JPEG': 1024}

ImageCompressTransformation

ImageCompressTransformation(
    init_quality: int = 70,
    max_size_bytes: Optional[int] = None,
    compression_ratio: Optional[float] = None,
    mode: Optional[str] = None,
    format: str = "JPEG",
    last_channel: bool = True,
    header_bytes: Optional[int] = None,
)

Bases: DataTransformation

Initialize JPEG compression transformation. Args: init_quality: Initial JPEG quality setting (1-100). max_size_bytes: Maximum allowed size of compressed JPEG in bytes. compression_ratio: Size budget as H * W * C // compression_ratio + 1 + header_bytes bytes (defaults to a conservative ratio for format derived from init_quality). mode: Optional mode for PIL Image (e.g., "RGB", "L"). If None, inferred from input. format: Image format to use for compression (default "JPEG"). See https://pillow.readthedocs.io/en/stable/handbook/image-file-formats.html for options. last_channel: Whether images are (..., H, W, C) (True) or channel-less (..., H, W) (False). header_bytes: Fixed allowance added to ratio-based budgets for the per-image format overhead, which dominates for small images. Defaults to a conservative value for format; ignored when max_size_bytes is given.

Methods:

Name Description
validate_source_space
get_uint8_dtype
get_target_space_from_source
encode_to_size

Encode an image (H, W, 3) or (H, W, 1) as JPEG bytes,

transform
direction_inverse
serialize
deserialize_from

Attributes:

Name Type Description
has_inverse
init_quality
max_size_bytes
compression_ratio
mode
format
last_channel
header_bytes

has_inverse class-attribute instance-attribute

has_inverse = True

init_quality instance-attribute

init_quality = init_quality

max_size_bytes instance-attribute

max_size_bytes = max_size_bytes

compression_ratio instance-attribute

compression_ratio = (
    compression_ratio
    if compression_ratio is not None
    else CONSERVATIVE_COMPRESSION_RATIOS.get(format, None)(init_quality)
    if max_size_bytes is None
    else None
)

mode instance-attribute

mode = mode

format instance-attribute

format = format

last_channel instance-attribute

last_channel = last_channel

header_bytes instance-attribute

header_bytes = (
    header_bytes
    if header_bytes is not None
    else CONSERVATIVE_HEADER_BYTES.get(format, 0)
)

validate_source_space

validate_source_space(
    source_space: Space[Any, BDeviceType, BDtypeType, BRNGType],
) -> None

get_uint8_dtype staticmethod

get_uint8_dtype(
    backend: ComputeBackend[BArrayType, BDeviceType, BDtypeType, BRNGType],
) -> BDtypeType

get_target_space_from_source

get_target_space_from_source(source_space)

encode_to_size

encode_to_size(img_array, max_bytes, min_quality=20, mode=None)

Encode an image (H, W, 3) or (H, W, 1) as JPEG bytes, reducing quality until <= max_bytes.

Parameters:

Name Type Description Default
img_array

np.ndarray, uint8, shape (H, W, 3) RGB or (H, W, 1) grayscale

required
max_bytes

maximum allowed size of JPEG file

required
min_quality

minimum JPEG quality before giving up (capped at init_quality)

20

Returns:

Type Description

jpeg_bytes (bytes), final_quality (int)

Raises:

Type Description
ValueError

if the image does not fit in max_bytes even at min_quality.

transform

transform(source_space, data)

direction_inverse

direction_inverse(source_space=None)

serialize

serialize(
    source_space: Optional[Space[Any, BDeviceType, BDtypeType, BRNGType]] = None,
) -> Dict[str, Any]

deserialize_from classmethod

deserialize_from(
    json_data: Dict[str, Any],
    source_space: Optional[Space[Any, BDeviceType, BDtypeType, BRNGType]] = None,
)

ImageDecompressTransformation

ImageDecompressTransformation(
    target_height: int,
    target_width: int,
    target_channels: Optional[int] = 3,
    mode: Optional[str] = None,
    format: Optional[str] = None,
)

Bases: DataTransformation

Initialize JPEG decompression transformation. Args: target_height: Height of the decompressed image. target_width: Width of the decompressed image. mode: Optional mode for PIL Image (e.g., "RGB", "L"). If None, inferred from input. format: Image format to use for decompression (default None, which will try everything). See https://pillow.readthedocs.io/en/stable/handbook/image-file-formats.html for options.

Methods:

Name Description
validate_source_space
get_uint8_dtype
get_target_space_from_source
decode_bytes

Decode JPEG bytes to an image array (H, W, 3).

transform
direction_inverse
serialize
deserialize_from

Attributes:

Name Type Description
has_inverse
target_height
target_width
target_channels
mode
format

has_inverse class-attribute instance-attribute

has_inverse = True

target_height instance-attribute

target_height = target_height

target_width instance-attribute

target_width = target_width

target_channels instance-attribute

target_channels = target_channels

mode instance-attribute

mode = mode

format instance-attribute

format = format

validate_source_space staticmethod

validate_source_space(
    source_space: Space[Any, BDeviceType, BDtypeType, BRNGType],
) -> None

get_uint8_dtype staticmethod

get_uint8_dtype(backend)

get_target_space_from_source

get_target_space_from_source(source_space)

decode_bytes

decode_bytes(jpeg_bytes: bytes, mode=None)

Decode JPEG bytes to an image array (H, W, 3).

Parameters:

Name Type Description Default
jpeg_bytes bytes

bytes of JPEG image

required

Returns:

Name Type Description
img_array

np.ndarray, uint8, shape (H, W, 3)

transform

transform(source_space, data)

direction_inverse

direction_inverse(source_space=None)

serialize

serialize(
    source_space: Optional[Space[Any, BDeviceType, BDtypeType, BRNGType]] = None,
) -> Dict[str, Any]

deserialize_from classmethod

deserialize_from(
    json_data: Dict[str, Any],
    source_space: Optional[Space[Any, BDeviceType, BDtypeType, BRNGType]] = None,
)

get_jpeg_compression_ratio

get_jpeg_compression_ratio(init_quality: int) -> int