trtutils.image.kernels module

CUDA kernel implementations for various preprocessing functions.

:attribute:`SCALE_SWAP_TRANSPOSE`

Rescales an image, swaps channels, and transposes HWC -> CHW

Type:

tuple[Path, str]

:attribute:`SST_FAST`

Rescales an image, swaps channels, and transposes HWC -> CHW (float32 output)

Type:

tuple[Path, str]

:attribute:`SST_FAST_F16`

Rescales an image, swaps channels, and transposes HWC -> CHW (float16 output)

Type:

tuple[Path, str]

:attribute:`LETTERBOX_RESIZE`

Resizes an image using the letterbox method.

Type:

tuple[Path, str]

:attribute:`LINEAR_RESIZE`

Resizes and image using bilinear interpolation.

Type:

tuple[Path, str]

:attribute:`IMAGENET_SST`

ImageNet normalization with scale/swap/transpose (float32 output)

Type:

tuple[Path, str]

:attribute:`IMAGENET_SST_F16`

ImageNet normalization with scale/swap/transpose (float16 output)

Type:

tuple[Path, str]