PreprocessRuntimeMeta Struct

Per-buffer preprocessing context — the inverse-transform breadcrumb trail. More...

Declaration

struct simaai::neat::PreprocessRuntimeMeta { ... }

Included Headers

#include <TensorCore.h>

Public Member Functions Index

bool	has_axis_perm () const noexcept
	True iff axis_perm is non-empty (a layout permutation was recorded). More...
bool	has_roi_list () const noexcept
	True iff runtime ROI-list metadata is intentionally present. More...

Public Member Attributes Index

int	original_width = 0
	Width of the source frame in pixels before preprocess. More...
int	original_height = 0
	Height of the source frame in pixels before preprocess. More...
int	resized_width = 0
	Width after the resize step (before any padding). More...
int	resized_height = 0
	Height after the resize step (before any padding). More...
int	scaled_width = 0
	Width after scaling (resize × additional scale factor). More...
int	scaled_height = 0
	Height after scaling (resize × additional scale factor). More...
int	pad_left = 0
	Letterbox padding added on the left edge, in pixels. More...
int	pad_right = 0
	Letterbox padding added on the right edge, in pixels. More...
int	pad_top = 0
	Letterbox padding added on the top edge, in pixels. More...
int	pad_bottom = 0
	Letterbox padding added on the bottom edge, in pixels. More...
std::string	resize_mode
	Resize policy token (e.g., "letterbox", "stretch"). More...
std::string	color_in
	Input color format token (e.g., "bgr", "nv12"). More...
std::string	color_out
	Output color format token after color-convert. More...
std::vector< int >	axis_perm
	Axis permutation applied by preprocess layout_convert, if any. More...
bool	normalize = false
	True if a normalize step (mean/scale) was applied. More...
bool	quantize = false
	True if an INT8/INT16 quantize step was applied. More...
bool	tessellate = false
	True if a tessellate step (tile-block layout) was applied. More...
double	affine_m00 = 1.0
	2×3 affine matrix element (row 0, col 0): x scale. More...
double	affine_m01 = 0.0
	2×3 affine matrix element (row 0, col 1): x shear. More...
double	affine_m02 = 0.0
	2×3 affine matrix element (row 0, col 2): x translation. More...
double	affine_m10 = 0.0
	2×3 affine matrix element (row 1, col 0): y shear. More...
double	affine_m11 = 1.0
	2×3 affine matrix element (row 1, col 1): y scale. More...
double	affine_m12 = 0.0
	2×3 affine matrix element (row 1, col 2): y translation. More...
double	affine_scale_x = 1.0
	Scalar X scale factor from original to model coordinates. More...
double	affine_scale_y = 1.0
	Scalar Y scale factor from original to model coordinates. More...
double	affine_offset_x = 0.0
	Scalar X offset (typically pad_left) in model coordinates. More...
double	affine_offset_y = 0.0
	Scalar Y offset (typically pad_top) in model coordinates. More...
bool	roi_list_enabled = false
	True when runtime ROI-list metadata is intentionally present. More...
std::vector< PreprocessRoi >	rois
	Runtime ROI windows in output-slot order. More...
int	roi_input_batch_size = 0
	Number of source images available to batch_index. More...
int	roi_source_width = 0
	Width of each source image before ROI extraction. More...
int	roi_source_height = 0
	Height of each source image before ROI extraction. More...
int	roi_source_stride_bytes = 0
	Source row stride in bytes, if known. More...
int	roi_pad_value = 0
	Pad value used for out-of-frame ROI pixels. More...
int	roi_capacity = 0
	Output-slot capacity reserved for this ROI-list dispatch. More...
int	roi_valid_count = 0
	Number of ROI output slots containing valid data. More...
int	roi_input_count = 0
	Number of ROIs requested before runtime filtering. More...
int	roi_dropped_invalid = 0
	ROIs dropped because their descriptor was invalid. More...
int	roi_dropped_overflow = 0
	ROIs dropped because they exceeded output capacity. More...
std::vector< PreprocessAffine >	roi_affines
	Per-valid-ROI affines in output-slot order, when available. More...

Description

Per-buffer preprocessing context — the inverse-transform breadcrumb trail.

When the framework's preprocess stage resizes/letterboxes/normalizes an input image, it records what it did into this struct and attaches it as Tensor::semantic.preprocess. The downstream BoxDecode (and any user code) uses these fields to map model-space coordinates (e.g., detected boxes in 640×640) back to original image coordinates.

The affine_* fields encode the full 2×3 affine transform from original image coordinates to model input coordinates; invert this matrix to map detections back to the source frame.

Definition at line 319 of file TensorCore.h.

Public Member Functions

has_axis_perm()

bool simaai::neat::PreprocessRuntimeMeta::has_axis_perm ()

inline noexcept

True iff axis_perm is non-empty (a layout permutation was recorded).

Definition at line 368 of file TensorCore.h.

has_roi_list()

bool simaai::neat::PreprocessRuntimeMeta::has_roi_list ()

inline noexcept

True iff runtime ROI-list metadata is intentionally present.

Definition at line 373 of file TensorCore.h.

Public Member Attributes

affine_m00

double simaai::neat::PreprocessRuntimeMeta::affine_m00 = 1.0

2×3 affine matrix element (row 0, col 0): x scale.

Definition at line 341 of file TensorCore.h.

affine_m01

double simaai::neat::PreprocessRuntimeMeta::affine_m01 = 0.0

2×3 affine matrix element (row 0, col 1): x shear.

Definition at line 342 of file TensorCore.h.

affine_m02

double simaai::neat::PreprocessRuntimeMeta::affine_m02 = 0.0

2×3 affine matrix element (row 0, col 2): x translation.

Definition at line 343 of file TensorCore.h.

affine_m10

double simaai::neat::PreprocessRuntimeMeta::affine_m10 = 0.0

2×3 affine matrix element (row 1, col 0): y shear.

Definition at line 344 of file TensorCore.h.

affine_m11

double simaai::neat::PreprocessRuntimeMeta::affine_m11 = 1.0

2×3 affine matrix element (row 1, col 1): y scale.

Definition at line 345 of file TensorCore.h.

affine_m12

double simaai::neat::PreprocessRuntimeMeta::affine_m12 = 0.0

2×3 affine matrix element (row 1, col 2): y translation.

Definition at line 346 of file TensorCore.h.

affine_offset_x

double simaai::neat::PreprocessRuntimeMeta::affine_offset_x = 0.0

Scalar X offset (typically pad_left) in model coordinates.

Definition at line 349 of file TensorCore.h.

affine_offset_y

double simaai::neat::PreprocessRuntimeMeta::affine_offset_y = 0.0

Scalar Y offset (typically pad_top) in model coordinates.

Definition at line 350 of file TensorCore.h.

affine_scale_x

double simaai::neat::PreprocessRuntimeMeta::affine_scale_x = 1.0

Scalar X scale factor from original to model coordinates.

Definition at line 347 of file TensorCore.h.

affine_scale_y

double simaai::neat::PreprocessRuntimeMeta::affine_scale_y = 1.0

Scalar Y scale factor from original to model coordinates.

Definition at line 348 of file TensorCore.h.

axis_perm

std::vector<int> simaai::neat::PreprocessRuntimeMeta::axis_perm

Axis permutation applied by preprocess layout_convert, if any.

Definition at line 335 of file TensorCore.h.

color_in

std::string simaai::neat::PreprocessRuntimeMeta::color_in

Input color format token (e.g., "bgr", "nv12").

Definition at line 332 of file TensorCore.h.

color_out

std::string simaai::neat::PreprocessRuntimeMeta::color_out

Output color format token after color-convert.

Definition at line 333 of file TensorCore.h.

normalize

bool simaai::neat::PreprocessRuntimeMeta::normalize = false

True if a normalize step (mean/scale) was applied.

Definition at line 337 of file TensorCore.h.

original_height

int simaai::neat::PreprocessRuntimeMeta::original_height = 0

Height of the source frame in pixels before preprocess.

Definition at line 321 of file TensorCore.h.

original_width

int simaai::neat::PreprocessRuntimeMeta::original_width = 0

Width of the source frame in pixels before preprocess.

Definition at line 320 of file TensorCore.h.

pad_bottom

int simaai::neat::PreprocessRuntimeMeta::pad_bottom = 0

Letterbox padding added on the bottom edge, in pixels.

Definition at line 329 of file TensorCore.h.

pad_left

int simaai::neat::PreprocessRuntimeMeta::pad_left = 0

Letterbox padding added on the left edge, in pixels.

Definition at line 326 of file TensorCore.h.

pad_right

int simaai::neat::PreprocessRuntimeMeta::pad_right = 0

Letterbox padding added on the right edge, in pixels.

Definition at line 327 of file TensorCore.h.

pad_top

int simaai::neat::PreprocessRuntimeMeta::pad_top = 0

Letterbox padding added on the top edge, in pixels.

Definition at line 328 of file TensorCore.h.

quantize

bool simaai::neat::PreprocessRuntimeMeta::quantize = false

True if an INT8/INT16 quantize step was applied.

Definition at line 338 of file TensorCore.h.

resize_mode

std::string simaai::neat::PreprocessRuntimeMeta::resize_mode

Resize policy token (e.g., "letterbox", "stretch").

Definition at line 331 of file TensorCore.h.

resized_height

int simaai::neat::PreprocessRuntimeMeta::resized_height = 0

Height after the resize step (before any padding).

Definition at line 323 of file TensorCore.h.

resized_width

int simaai::neat::PreprocessRuntimeMeta::resized_width = 0

Width after the resize step (before any padding).

Definition at line 322 of file TensorCore.h.

roi_affines

std::vector<PreprocessAffine> simaai::neat::PreprocessRuntimeMeta::roi_affines

Per-valid-ROI affines in output-slot order, when available.

Definition at line 365 of file TensorCore.h.

roi_capacity

int simaai::neat::PreprocessRuntimeMeta::roi_capacity = 0

Output-slot capacity reserved for this ROI-list dispatch.

Definition at line 359 of file TensorCore.h.

roi_dropped_invalid

int simaai::neat::PreprocessRuntimeMeta::roi_dropped_invalid = 0

ROIs dropped because their descriptor was invalid.

Definition at line 362 of file TensorCore.h.

roi_dropped_overflow

int simaai::neat::PreprocessRuntimeMeta::roi_dropped_overflow = 0

ROIs dropped because they exceeded output capacity.

Definition at line 363 of file TensorCore.h.

roi_input_batch_size

int simaai::neat::PreprocessRuntimeMeta::roi_input_batch_size = 0

Number of source images available to batch_index.

Definition at line 354 of file TensorCore.h.

roi_input_count

int simaai::neat::PreprocessRuntimeMeta::roi_input_count = 0

Number of ROIs requested before runtime filtering.

Definition at line 361 of file TensorCore.h.

roi_list_enabled

bool simaai::neat::PreprocessRuntimeMeta::roi_list_enabled = false

True when runtime ROI-list metadata is intentionally present.

Definition at line 352 of file TensorCore.h.

roi_pad_value

int simaai::neat::PreprocessRuntimeMeta::roi_pad_value = 0

Pad value used for out-of-frame ROI pixels.

Definition at line 358 of file TensorCore.h.

roi_source_height

int simaai::neat::PreprocessRuntimeMeta::roi_source_height = 0

Height of each source image before ROI extraction.

Definition at line 356 of file TensorCore.h.

roi_source_stride_bytes

int simaai::neat::PreprocessRuntimeMeta::roi_source_stride_bytes = 0

Source row stride in bytes, if known.

Definition at line 357 of file TensorCore.h.

roi_source_width

int simaai::neat::PreprocessRuntimeMeta::roi_source_width = 0

Width of each source image before ROI extraction.

Definition at line 355 of file TensorCore.h.

roi_valid_count

int simaai::neat::PreprocessRuntimeMeta::roi_valid_count = 0

Number of ROI output slots containing valid data.

Definition at line 360 of file TensorCore.h.

rois

std::vector<PreprocessRoi> simaai::neat::PreprocessRuntimeMeta::rois

Runtime ROI windows in output-slot order.

Definition at line 353 of file TensorCore.h.

scaled_height

int simaai::neat::PreprocessRuntimeMeta::scaled_height = 0

Height after scaling (resize × additional scale factor).

Definition at line 325 of file TensorCore.h.

scaled_width

int simaai::neat::PreprocessRuntimeMeta::scaled_width = 0

Width after scaling (resize × additional scale factor).

Definition at line 324 of file TensorCore.h.

tessellate

bool simaai::neat::PreprocessRuntimeMeta::tessellate = false

True if a tessellate step (tile-block layout) was applied.

Definition at line 339 of file TensorCore.h.

---

The documentation for this struct was generated from the following file:

• TensorCore.h

---

Generated via doxygen2docusaurus 2.0.0 by Doxygen 1.9.8.