Class holding an externally computed disparity or depth map, to be ingested with sl::Camera::ingestCustomDepth(). More...

Attributes
sl::Mat	map
	Disparity or depth map, sl::MAT_TYPE::F32_C1, in CPU or GPU memory. More...

CUSTOM_DEPTH_FORMAT	format = CUSTOM_DEPTH_FORMAT::DISPARITY
	Content type of map. More...

float	scale = 1.f
	Multiplier applied to each value of map before interpretation. More...

sl::Mat	confidence
	Optional confidence map, sl::MAT_TYPE::F32_C1, same resolution as map, CPU or GPU memory. More...

CUSTOM_CONFIDENCE_CONVENTION	confidence_convention = CUSTOM_CONFIDENCE_CONVENTION::PROBABILITY
	Value convention of confidence. Only consulted when confidence is set. More...

sl::Timestamp	timestamp = 0
	Timestamp of the frame the map was computed from, i.e. sl::Camera::getTimestamp(sl::TIME_REFERENCE::IMAGE) after sl::Camera::read(). More...

Detailed Description

Class holding an externally computed disparity or depth map, to be ingested with sl::Camera::ingestCustomDepth().

Warning: Experimental feature: downstream modules may not be fully robust to arbitrary custom depth/confidence input and can produce unexpected behavior or reduced accuracy.

Requires sl::InitParameters::depth_mode set to sl::DEPTH_MODE::CUSTOM. The expected sequence per frame is: sl::Camera::read(), retrieve the rectified images, compute the map externally, ingestCustomDepth(), then sl::Camera::grab(). The SDK resamples the map to its internal depth resolution (disparity values are rescaled accordingly), so the map can be provided at any resolution, typically the network output resolution.

Variables

◆ map

sl::Mat map

Disparity or depth map, sl::MAT_TYPE::F32_C1, in CPU or GPU memory.

The data is consumed during the call: the sl::Mat can be reused or freed as soon as ingestCustomDepth() returns.

◆ format

CUSTOM_DEPTH_FORMAT format = CUSTOM_DEPTH_FORMAT::DISPARITY

Content type of map.

◆ scale

float scale = 1.f

Multiplier applied to each value of map before interpretation.

Lets normalized or differently-conventioned outputs be ingested without an extra pass: e.g. set it to the maximum disparity if the network outputs disparity normalized to [0,1], to -1 if it outputs negative disparities (sl::MEASURE::DISPARITY convention), or to a unit factor if the depth unit differs from sl::InitParameters::coordinate_units. Default: 1

◆ confidence

sl::Mat confidence

Optional confidence map, sl::MAT_TYPE::F32_C1, same resolution as map, CPU or GPU memory.

Strongly recommended when available. If left unset, a confidence map is derived from the validity of map (valid pixels fully trusted, invalid pixels rejected); sl::RuntimeParameters::confidence_threshold then acts as a per-pixel on/off filter. Invalid map pixels are always given the worst confidence, whatever this map contains.

◆ confidence_convention

CUSTOM_CONFIDENCE_CONVENTION confidence_convention = CUSTOM_CONFIDENCE_CONVENTION::PROBABILITY

Value convention of confidence. Only consulted when confidence is set.

◆ timestamp

sl::Timestamp timestamp = 0

Timestamp of the frame the map was computed from, i.e. sl::Camera::getTimestamp(sl::TIME_REFERENCE::IMAGE) after sl::Camera::read().

Used to detect frame mismatches (a warning is logged if it differs from the current frame). Optional: 0 disables the check.

Attributes