Face Engine SDK 5.25.0
A face detection, recognition and tracking engine.
Loading...
Searching...
No Matches
fsdk::IDescriptorExtractor Struct Referenceabstract

Descriptor extractor interface. More...

#include <IDescriptor.h>

Inheritance diagram for fsdk::IDescriptorExtractor:
fsdk::IRefCounted

Public Types

using FutureResult = vlc::future<float>
 Common aliases for BestShotQuality asynchronous interface.
 

Public Member Functions

virtual ResultValue< FSDKError, float > extractFromWarpedImage (const Image &warp, IDescriptor *descriptor) const noexcept=0
 Extract descriptor from a warped image.
 
virtual ResultValue< FSDKError, float > extractFromWarpedImageBatch (Span< const Image > warps, IDescriptorBatch *descriptorBatch, IDescriptor *aggregation, Span< float > garbageScoreBatch) const noexcept=0
 Extract batch of descriptors from a batch of images and perform aggregation.
 
virtual Result< FSDKErrorextractFromWarpedImageBatch (Span< const Image > warps, IDescriptorBatch *descriptorBatch, Span< float > garbageScoreBatch) const noexcept=0
 Extract batch of descriptors from a batch of images.
 
virtual Result< FSDKErrorvalidate (Span< const Image > warps, Span< Result< FSDKError > > errors) const noexcept=0
 Validate input of multiple frames in a single function call.
 
virtual uint32_t getModelVersion () const noexcept=0
 Get algorithm model version this extractor works with.
 
virtual DescriptorType getDescriptorType () const noexcept=0
 Get type of descriptor this extractor works with.
 
virtual FutureResult extractFromWarpedImageBatchAsync (Span< const Image > warps, IDescriptorBatch *descriptorBatch, IDescriptor *aggregation, Span< float > garbageScoreBatch) const =0
 Asynchronously extract batch of descriptors from a batch of images.
 
- Public Member Functions inherited from fsdk::IRefCounted
virtual int32_t retain () noexcept=0
 Increase strong reference count.
 
virtual int32_t retainLocked () noexcept=0
 Increase strong reference count thread safely.
 
virtual int32_t release () noexcept=0
 Decrease strong reference count.
 
virtual int32_t getRefCount () const noexcept=0
 Get actual strong reference count.
 
virtual int32_t retainWeak () noexcept=0
 Increase weak reference count.
 
virtual int32_t releaseWeak () noexcept=0
 Decrease weak reference count.
 
virtual int32_t getWeakRefCount () const noexcept=0
 Get actual weak reference count.
 

Detailed Description

Descriptor extractor interface.

Extracts face descriptors from images. The descriptors can be later used for face matching.

Member Function Documentation

◆ extractFromWarpedImage()

virtual ResultValue< FSDKError, float > fsdk::IDescriptorExtractor::extractFromWarpedImage ( const Image & warp,
IDescriptor * descriptor ) const
pure virtualnoexcept

Extract descriptor from a warped image.

Parameters
[in]warpimage with warped face or human warp.
Note
Warp should be a valid 250x250 image in R8G8B8 format for DT_FACE descriptor type.
Warp should be a valid 128x256 image in R8G8B8 format for DT_HUMAN descriptor type.
Parameters
[out]descriptordescriptor to fill with data.
Returns
ResultValue with error code and score of descriptor normalized in range [0, 1] 1 - face on the input warp; 0 - garbage on the input warp.
Note
human descriptor does not support garbage score, 1.0 will be returned.
See also
Image, IDescriptor, ResultValue and FSDKError for details.
Note
warp format must be R8G8B8,
See also
Format.

◆ extractFromWarpedImageBatch() [1/2]

virtual ResultValue< FSDKError, float > fsdk::IDescriptorExtractor::extractFromWarpedImageBatch ( Span< const Image > warps,
IDescriptorBatch * descriptorBatch,
IDescriptor * aggregation,
Span< float > garbageScoreBatch ) const
pure virtualnoexcept

Extract batch of descriptors from a batch of images and perform aggregation.

Parameters
[in]warpsspan of images with warped faces or human warps.
Note
Warps should be in R8G8B8 format, with size 250x250 for DT_FACE descriptor type.
Warps should be in R8G8B8 format, with size 128x256 for DT_HUMAN descriptor type.
Parameters
[out]descriptorBatchdescriptor batch to fill with data.
Note
DT_HUMAN descriptor does not support garbage score.
Parameters
[out]aggregationdescriptor with aggregation based on descriptor batch.
[out]garbageScoreBatchspan of descriptor scores normalized in range [0, 1] 1 - face on the input warp; 0 - garbage on the input warp.
Note
human descriptor does not support garbage score, 1.0 will be returned.
Returns
ResultValue with error code and aggregated garbage score.
See also
Span, Image, IDescriptor, IDescriptorBatch, ResultValue and FSDKError for details.
Note
warps format must be R8G8B8,
See also
Format.
Note
all spans should be based on user owned continuous collections.
all spans should be equal size.

◆ extractFromWarpedImageBatch() [2/2]

virtual Result< FSDKError > fsdk::IDescriptorExtractor::extractFromWarpedImageBatch ( Span< const Image > warps,
IDescriptorBatch * descriptorBatch,
Span< float > garbageScoreBatch ) const
pure virtualnoexcept

Extract batch of descriptors from a batch of images.

Parameters
[in]warpsspan of images with warped faces or human warps.
Note
Warps should be in R8G8B8 format, with size 250x250 for DT_FACE descriptor type.
Warps should be in R8G8B8 format, with size 128x256 for DT_HUMAN descriptor type.
Parameters
[out]descriptorBatchdescriptor batch to fill with data.
[out]garbageScoreBatchspan of descriptor scores normalized in range [0, 1] 1 - face on the input warp; 0 - garbage on the input warp. DT_HUMAN descriptor does not support garbage score. In a case of DT_HUMAN descriptor, you'll get batch filled by 1.0.
Returns
Result with error code.
See also
Span, Image, IDescriptorBatch, Result and FSDKError for details.
Note
warps format must be R8G8B8,
See also
Format.
Note
all spans should be based on user owned continuous collections.
all spans should be equal size.

◆ extractFromWarpedImageBatchAsync()

virtual FutureResult fsdk::IDescriptorExtractor::extractFromWarpedImageBatchAsync ( Span< const Image > warps,
IDescriptorBatch * descriptorBatch,
IDescriptor * aggregation,
Span< float > garbageScoreBatch ) const
pure virtual

Asynchronously extract batch of descriptors from a batch of images.

Parameters
[in]warpsspan of images with warped faces or human warps.
Note
Warps should be in R8G8B8 format, with size 250x250 for DT_FACE descriptor type.
Warps should be in R8G8B8 format, with size 128x256 for DT_HUMAN descriptor type.
Parameters
[out]descriptorBatchdescriptor batch to fill with data.
[out]aggregationdescriptor with aggregation based on descriptor batch.
[out]garbageScoreBatchspan of descriptor scores normalized in range [0, 1] 1 - face on the input warp; 0 - garbage on the input warp. DT_HUMAN descriptor does not support garbage score. In a case of DT_HUMAN descriptor, you'll get batch filled by 1.0.
Returns
Result with error code and aggregated garbage score.
See also
Span, Image, IDescriptorBatch, Result and FSDKError for details.
Note
warps format must be R8G8B8,
See also
Format.
Note
all spans should be based on user owned continuous collections.
all spans should be equal size.
this method is experimental and interface may be changed in the future.
this method is not marked as noexcept and may throw an exception.

◆ getDescriptorType()

virtual DescriptorType fsdk::IDescriptorExtractor::getDescriptorType ( ) const
pure virtualnoexcept

Get type of descriptor this extractor works with.

Returns
type as enum
See also
DescriptorType.

◆ getModelVersion()

virtual uint32_t fsdk::IDescriptorExtractor::getModelVersion ( ) const
pure virtualnoexcept

Get algorithm model version this extractor works with.

Returns
Version as integral number.

◆ validate()

virtual Result< FSDKError > fsdk::IDescriptorExtractor::validate ( Span< const Image > warps,
Span< Result< FSDKError > > errors ) const
pure virtualnoexcept

Validate input of multiple frames in a single function call.

Parameters
[in]warpsspan of images with warped faces or human warps.
[out]errorsoutput span of errors for each image.
Returns
Result with error code.
See also
Span, Image, Result and FSDKError for details.
Note
warps format must be R8G8B8,
See also
Format.
Note
all spans should be based on user owned continuous collections.
all spans should be equal size.

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