The MyCaffe.common namespace contains common MyCaffe classes. More...

Classes
class	ActionStateArgs
	The ActionStateArgs are sent to the DoWork event when fired from the InternalThreadEntry. More...

class	BackwardArgs
	The BackwardArgs are passed to the OnBackward event of the EventLayer. More...

class	BatchInput
	The BatchInput class stores the mini-batch index and input data. More...

class	BBoxUtility
	The BBox class processes the NormalizedBBox data used with SSD. More...

class	BeamSearch
	The BeamSearch uses the softmax output from the network and continually runs the net on each output (using the output as input) until the end of token is reached. The beam-search tree is returned. More...

class	Blob
	The Blob is the main holder of data that moves through the Layers of the Net. More...

class	BlobCollection
	The BlobCollection contains a list of Blobs. More...

class	BlobDebugInformation
	The BlobDebugInformation describes debug information relating to a given Blob in a given Layer. More...

class	BlobName
	The BlobName class is used to build unique blob names. More...

class	BlockingQueue
	The BlockingQueue is used for synchronized Queue operations. More...

class	Cache
	The Cache class is used to cache blobs over time. More...

class	ComputeGraph
	The ComputeGraph class provides a simple computation graph of operations used in a forward pass that are stored in an array on each call and then unwound with calls that calculate the gradients on the backward pass. More...

class	CudaDnn
	The CudaDnn object is the main interface to the Low-Level Cuda C++ DLL. More...

class	CudaDnnMemoryTracker
	The CudaDnnMemoryTracker is used for diagnostics in that it helps estimate the amount of memory that a Net will use. More...

class	CustomForwardBackArgs
	The CustomForwardBackArgs provide the arguments to the OnCustomForwardBack event within the Solver Step function. More...

class	DebugInformation
	The DebugInformation contains information used to help debug the Layers of a Net while it is training. More...

class	ForwardArgs
	The ForwardArgs are passed to the OnForward event of the EventLayer. More...

class	GetBytesArgs
	The GetBytesArgs is passed along to the SnapshotArgs::OnGetWeights and SnapshotArgs::OnGetState events. More...

class	GetConversionBlobArgs
	The GetConversionBlobArgs are passed to the Layer::OnGetConversionBlobs event which fires each time a blob needs to be converted form half to base or back. More...

class	GetIterationArgs
	The GetIterationArgs is sent bubbled up to the solver when a layer needs to know the curret training iteration. More...

class	GetWorkBlobArgs
	The GetWorkBlobArgs are passed to the Layer::OnGetWorkBlob event which is supported for debugging only. More...

class	GPUParams
	The GPUParams contains the connection to the low-level Cuda, and the stream associated with this instance. More...

class	GradientsReadyArgs
	The GradientsReadyArgs is sent to the Solver::OnGradientsReady event which fires at the end of each Solver::Step. More...

class	HostBuffer
	The HostBuffer helps manage host memory, often used when implementing CPU versions of a function or layer. More...

class	InternalThread
	The InternalThread manages an internal thread used for Parallel and data collection operations. More...

interface	IXDebugData
	The IXDebugData interface is implemented by the DebugLayer to give access to the debug information managed by the layer. More...

interface	IXMyCaffe
	The IXMyCaffe interface contains functions used to perform MyCaffe operations that work with the MyCaffeImageDatabase. More...

interface	IXMyCaffeExtension
	The IXMyCaffeExtension interface allows for easy extension management of the low-level software that interacts directly with CUDA. More...

interface	IXMyCaffeNoDb
	The IXMyCaffeNoDb interface contains functions used to perform MyCaffe operations that run in a light-weight manner without the MyCaffeImageDatabase. More...

interface	IXMyCaffeState
	The IXMyCaffeState interface contains functions related to the MyCaffeComponent state. More...

interface	IXPersist
	The IXPersist interface is used by the CaffeControl to load and save weights. More...

class	LayerDebugInformation
	The LayerDebugInformation describes debug information relating to a given Layer in the Net. More...

class	NCCL
	The NCCL class manages the multi-GPU operations using the low-level NCCL functionality provided by the low-level Cuda Dnn DLL. More...

class	Net
	Connects Layer's together into a direct acrylic graph (DAG) specified by a NetParameter More...

class	NumpyFile
	The NumpyFile reads data from a numpy file in the base type specified. More...

class	Params
	The Params contains the base parameters used in multi-GPU training. More...

class	PersistCaffe
	The PersistCaffe class is used to load and save weight files in the .caffemodel format. More...

class	Property
	The Property class stores both a numeric and text value. More...

class	PropertyTree
	The PropertyTree class implements a simple property tree similar to the ptree in Boost. More...

class	ResultCollection
	The ResultCollection contains the result of a given CaffeControl::Run. More...

class	SnapshotArgs
	The SnapshotArgs is sent to the Solver::OnSnapshot event which fires each time the Solver::Snapshot method is called. More...

class	SolverInfo
	The SolverInfo defines the user supplied arguments passed to each Worker. More...

class	SsdSampler
	The SsdSampler is used by the SSD algorithm to sample BBoxes. More...

class	SyncedMemory
	The SyncedMemory manages the low-level connection between the GPU and host memory. More...

class	TestArgs
	The TestArgs are passed to the Solver::OnTest event. More...

class	TestingIterationArgs
	Specifies the TestingIterationArgs sent to the Solver::OnTestingIteration, which is called at the end of a testing cycle. More...

class	TestResultArgs
	The TestResultArgs are passed to the Solver::OnTestResults event. More...

class	TrainingIterationArgs
	The TrainingIterationArgs is sent to the Solver::OnTrainingIteration event that fires at the end of a training cycle. More...

class	TransferInput
	The TransferInput class is used to transfer get and set input data. More...

class	WeightInfo
	The WeightInfo class describes the weights of a given weight set including the blob names and sizes of the weights. More...

class	Worker
	The Worker manages each 'non' root sover running, where each Worker operates on a different GPU. More...

class	WorkspaceArgs
	The WorkspaceArgs are passed to both the Layer::OnSetWorkspace and Layer::OnGetWorkspace events. More...

Enumerations
enum	DIR { FWD = 0 , BWD = 1 }
	Defines the direction of data flow. More...

enum	MEAN_ERROR { MSE = 1 , MAE = 2 }
	Defines the type of Mean Error to use. More...

enum	MATH_FUNCTION { NOP = 0 , ACOS = 1 , ACOSH = 2 , COS = 3 , COSH = 4 , ASIN = 10 , ASINH = 11 , SIN = 12 , SINH = 13 , ATAN = 20 , ATANH = 21 , TAN = 22 , TANH = 23 , CEIL = 30 , FLOOR = 31 , NEG = 32 , SIGN = 33 , SQRT = 34 }
	Defines the mathematical function to run. More...

enum	OP { MUL = 1 , DIV = 2 , ADD = 3 , SUB = 4 }
	Defines the operations performed by the channel_op function. More...

enum	DistanceMethod { HAMMING = 0 , EUCLIDEAN = 1 }
	Specifies the distance method used when calculating batch distances. More...

enum	PoolingMethod { MAX = 0 , AVE = 1 }
	Specifies the pooling method used by the cuDnn function SetPoolingDesc. More...

enum	DataType { DOUBLE , FLOAT }
	Specifies the base datatype corresponding the the template type 'T'. Currently, only More...

enum	DEVINIT { NONE = 0x0000 , CUBLAS = 0x0001 , CURAND = 0x0002 , SETSEED = 0x0004 }
	Specifies the initialization flags used when initializing CUDA. More...

enum	BATCHNORM_MODE { PER_ACTIVATION = 0 , SPATIAL = 1 , SPATIAL_PERSISTENT = 2 }
	Specifies the cuDnn batch norm mode to use. More...

enum	CONV_FWD_ALGO { NONE = -1 , IMPLICIT_GEMM = 0 , IMPLICIT_PRECOMP_GEMM = 1 , ALGO_GEMM = 2 , ALGO_DIRECT = 3 , ALGO_FFT = 4 , ALGO_FFT_TILING = 5 , ALGO_WINOGRAD = 6 , ALGO_WINOGRAD_NONFUSED = 7 }
	Specifies the cuDnn convolution forward algorithm to use. More...

enum	CONV_BWD_FILTER_ALGO { ALGO_0 = 0 , ALGO_1 = 1 , ALGO_FFT = 2 , ALGO_3 = 3 }
	Specifies the cuDnn convolution backward filter algorithm to use. More...

enum	CONV_BWD_DATA_ALGO { ALGO_0 = 0 , ALGO_1 = 1 , ALGO_FFT = 2 }
	Specifies the cuDnn convolution backward data algorithm to use. More...

enum	POOLING_METHOD { MAX = 0 , AVE = 1 , STO_TRAIN = 2 , STO_TEST = 3 }
	Specifies the pooling method to use when using the Caffe pooling (instead of the pooling from NVIDIA's cuDnn). More...

enum	RNN_MODE { RNN_RELU = 0 , RNN_TANH = 1 , LSTM = 2 , GRU = 3 }
	Specifies the RNN mode to use with the Recurrent Layer when using the cuDNN engine. More...

enum	RNN_BIAS_MODE { RNN_NO_BIAS = 0 , RNN_SINGLE_INP_BIAS = 1 , RNN_DOUBLE_BIAS = 2 , RNN_SINGLE_REC_BIAS = 3 }
	Specifies the RNN bias mode to use with the Recurrent Layer when using the cuDNN engine. More...

enum	RNN_DATALAYOUT { RNN_SEQ_MAJOR_UNPACKED = 0 , RNN_SEQ_MAJOR_PACKED = 1 , RNN_BATCH_MAJOR_UNPACKED = 2 }
	Specifies the RNN data layout of the data input. More...

enum	RNN_DIRECTION { RNN_UNIDIRECTIONAL , RNN_BIDIRECTIONAL }
	Specifies the RNN directional used. More...

enum	RNN_FILLER_TYPE { RNN_CONSTANT_FILLER , RNN_XAVIER_FILLER , RNN_GAUSSIAN_FILLER }
	Defines the filler types used to fill the RNN8 weights. More...

enum	DEVPROP { DEVICECOUNT = 1 , NAME = 2 , MULTIGPUBOARDGROUPID = 3 }
	Specifies certain device properties to query from Cuda. More...

enum	MEMTEST_TYPE { MOV_INV_8 = 1 }
	Specifies the memory test to perform. More...

enum	NCCL_REDUCTION_OP { SUM = 0 , PROD = 1 , MAX = 2 , MIN = 3 }
	Specifies the reduction operation to use with 'Nickel' NCCL. More...

enum	SSD_MINING_TYPE { NONE = 0 , MAX_NEGATIVE = 1 , HARD_EXAMPLE = 2 }
	Defines the mining type used during SSD cuda training. More...

enum	SSD_MATCH_TYPE { BIPARTITE , PER_PREDICTION }
	Defines the matching method used during SSD cuda training. More...

enum	SSD_CODE_TYPE { CORNER = 1 , CENTER_SIZE = 2 , CORNER_SIZE = 3 }
	Defines the encode/decode type used during SSD cuda training. More...

enum	SSD_CONF_LOSS_TYPE { SOFTMAX , LOGISTIC }
	Defines the confidence loss types used during SSD cuda training. More...

enum	SSD_LOC_LOSS_TYPE { L2 , SMOOTH_L1 }
	Defines the location loss types used during SSD cuda training. More...

enum	ORIENTATION { COL = 0 , ROW = 1 }
	Specifies the orientation of a matrix. More...

enum	TRANSPOSE_OPERATION { ADD = 0 , MUL = 1 , DIV = 2 }
	Specifies the type of operation to perform along with a matrix transposition. More...

enum	AGGREGATIONS { SUM = 0 , MAX = 1 , MIN = 2 }
	Specifies different aggregation operations. More...

enum	SOFTMAX_ALGORITHM { DEFAULT = 1 , FAST = 0 , ACCURATE = 1 , LOG = 2 }
	Specifies the SOFTMAX algorithm to use. More...

enum	SOFTMAX_MODE { INSTANCE , CHANNEL }
	Specifies the SOFTMAX mode to use. More...

enum	WEIGHT_TARGET { NONE , WEIGHTS , BIAS , BOTH }
	Defines the type of weight to target in re-initializations. More...

enum	BLOB_TYPE { UNKNOWN = 0x0000 , DATA = 0x0001 , IP_WEIGHT = 0x0002 , WEIGHT = 0x0004 , LOSS = 0x0008 , ACCURACY = 0x0010 , CLIP = 0x0020 , MULTIBBOX = 0x0040 , INTERNAL = 0x0080 , TARGET = 0x0100 , PREDICTION = 0x0200 , ATTENTION = 0x0400 }
	Defines the tpe of data held by a given Blob. More...

enum	TRAIN_STEP { NONE = 0x0000 , FORWARD = 0x0001 , BACKWARD = 0x0002 , BOTH = 0x0003 }
	Defines the training stepping method (if any). More...

Functions
delegate void	onSetWorkspace (object sender, WorkspaceArgs e)
	Delegate used to set the OnSetworkspace event. More...

delegate void	onGetWorkspace (object sender, WorkspaceArgs e)
	Delegate used to set the OnGetworkspace event. More...

Detailed Description

The MyCaffe.common namespace contains common MyCaffe classes.

Enumeration Type Documentation

◆ AGGREGATIONS

enum MyCaffe.common.AGGREGATIONS

Specifies different aggregation operations.

Enumerator
SUM	Sum the values.
MAX	Return the maximum value.
MIN	Return the minimum value.

Definition at line 681 of file CudaDnn.cs.

◆ BATCHNORM_MODE

enum MyCaffe.common.BATCHNORM_MODE

Specifies the cuDnn batch norm mode to use.

See also: NVIDIA cuDnn documenation for more details.

Enumerator
PER_ACTIVATION	Specifies to use the per-activation batch normalization mode.
SPATIAL	Specifies to use the spatial batch normalization mode.
SPATIAL_PERSISTENT	Specifies to use the spatial persistent batch normalization mode.

Definition at line 236 of file CudaDnn.cs.

◆ BLOB_TYPE

enum MyCaffe.common.BLOB_TYPE

Defines the tpe of data held by a given Blob.

Enumerator
UNKNOWN	The blob is an unknown type.
DATA	The Blob holds Data.
IP_WEIGHT	The Blob holds an inner product weight.
WEIGHT	The Blob holds a general weight.
LOSS	The Blob holds Loss Data.
ACCURACY	The Blob holds Accuracy Data.
CLIP	The blob holds Clip data.
MULTIBBOX	The blob holds multi-boundingbox data. The multi-box data ordering is as follows: [0] index of num. [1] label [2] score [3] bbox xmin [4] bbox ymin [5] bbox xmax [6] bbox ymax continues for each of the top 'n' bboxes output.
INTERNAL	The blob is an internal blob used within the layer.
TARGET	The blob contains target data.
PREDICTION	The blob contains prediction data.
ATTENTION	The blob contains attention scores.

Definition at line 61 of file Interfaces.cs.

◆ CONV_BWD_DATA_ALGO

enum MyCaffe.common.CONV_BWD_DATA_ALGO

Specifies the cuDnn convolution backward data algorithm to use.

See also: NVIDIA cuDnn documenation for more details.

Enumerator
ALGO_0	Specifies to use algorithm 0 - which is non-deterministic.
ALGO_1	Specifies to use algorithm 1.
ALGO_FFT	Specifies to use the fft algorithm.

Definition at line 330 of file CudaDnn.cs.

◆ CONV_BWD_FILTER_ALGO

enum MyCaffe.common.CONV_BWD_FILTER_ALGO

Specifies the cuDnn convolution backward filter algorithm to use.

See also: NVIDIA cuDnn documenation for more details.

Enumerator
ALGO_0	Specifies to use algorithm 0 - which is non-deterministic.
ALGO_1	Specifies to use algorithm 1.
ALGO_FFT	Specifies to use the fft algorithm.
ALGO_3	Specifies to use algorithm 0 with a workspace - which is non-deterministic.

Definition at line 304 of file CudaDnn.cs.

◆ CONV_FWD_ALGO

enum MyCaffe.common.CONV_FWD_ALGO

Specifies the cuDnn convolution forward algorithm to use.

See also: NVIDIA cuDnn documenation for more details.

Enumerator
NONE	Specifies to not use a forward algorithm.
IMPLICIT_GEMM	Specifies to use the implicit gemm algorithm.
IMPLICIT_PRECOMP_GEMM	Specifies to use the implicit pre-computation gemm algorithm.
ALGO_GEMM	Specifies to use the gemm algorithm.
ALGO_DIRECT	Specifies to use the direct algorithm.
ALGO_FFT	Specifies to use the fft algorithm.
ALGO_FFT_TILING	Specifies to use the fft tiling algorithm.
ALGO_WINOGRAD	Specifies to use the winograd algorithm.
ALGO_WINOGRAD_NONFUSED	Specifies to use the non-fused winograd algorithm.

Definition at line 258 of file CudaDnn.cs.

◆ DataType

enum MyCaffe.common.DataType

Specifies the base datatype corresponding the the template type 'T'. Currently, only

double and float types are supported.

Enumerator
DOUBLE	Specifies the double type.
FLOAT	Specifies the single type.

Definition at line 191 of file CudaDnn.cs.

◆ DEVINIT

enum MyCaffe.common.DEVINIT

Specifies the initialization flags used when initializing CUDA.

Enumerator
NONE	No flag specified.
CUBLAS	Initialize cuBlas. This should be initialized for cuBlas is used for many of the math operations.
CURAND	Initialize cuRand. This should be initialized for cuRand is used for most of the random operations.
SETSEED	Set the cuRand random number generator seed - typically only used when testing to ensure that random numbers are generated in a predictable ordering.

Definition at line 206 of file CudaDnn.cs.

◆ DEVPROP

enum MyCaffe.common.DEVPROP

Specifies certain device properties to query from Cuda.

Enumerator
DEVICECOUNT	Query the number of devices (gpu's) installed.
NAME	Query the name of a given GPU.
MULTIGPUBOARDGROUPID	Query a GPU board group ID.

Definition at line 476 of file CudaDnn.cs.

◆ DIR

enum MyCaffe.common.DIR

Defines the direction of data flow.

Enumerator
FWD	Specifies data is moving forward.
BWD	Specifies data is moving backward.

Definition at line 21 of file CudaDnn.cs.

◆ DistanceMethod

enum MyCaffe.common.DistanceMethod

Specifies the distance method used when calculating batch distances.

Enumerator
HAMMING	Specifies to calculate the hamming distance.
EUCLIDEAN	Specifies to calculate the euclidean distance.

Definition at line 158 of file CudaDnn.cs.

◆ MATH_FUNCTION

enum MyCaffe.common.MATH_FUNCTION

Defines the mathematical function to run.

Enumerator
NOP	Specifies to run a no operation.
ACOS	Specifies to run the acos function.
ACOSH	Specifies to run the acosh function.
COS	Specifies to run the cos function.
COSH	Specifies to run the cosh function.
ASIN	Specifies to run the asin function.
ASINH	Specifies to run the asinh function.
SIN	Specifies to run the sin function.
SINH	Specifies to run the sinh function.
ATAN	Specifies to run the atan function.
ATANH	Specifies to run the atanh function.
TAN	Specifies to run the tan function.
TANH	Specifies to run the tanh function.
CEIL	Specifies to run the ceil function.
FLOOR	Specifies to run the floor function.
NEG	Specifies to flip the sign of the inputs.
SIGN	Specifies to run the sign function.
SQRT	Specifies to run the sqrt function.

Definition at line 51 of file CudaDnn.cs.

◆ MEAN_ERROR

enum MyCaffe.common.MEAN_ERROR

Defines the type of Mean Error to use.

Enumerator
MSE	Mean Squared Error (MSE) $L(y, y\hat) = \frac{1}{N} \sum_{i=0}^{N} (y - \hat{y}{i})^2$ where $\hat{y}$ is the predicted value.
MAE	Mean Absolute Error (MAE) $L(y, y\hat) = \frac{1}{N} \sum_{i=0}^{N} \|y - \hat{y}{i}\|$ where $\hat{y}$ is the predicted value.

Definition at line 36 of file CudaDnn.cs.

◆ MEMTEST_TYPE

enum MyCaffe.common.MEMTEST_TYPE

Specifies the memory test to perform.

See also: CudaDnn::RunMemoryTest

Enumerator
MOV_INV_8	Specifies the mov-inv-8 test.

Definition at line 498 of file CudaDnn.cs.

◆ NCCL_REDUCTION_OP

enum MyCaffe.common.NCCL_REDUCTION_OP

Specifies the reduction operation to use with 'Nickel' NCCL.

See also: CudaDnn::NcclAllReduce

Enumerator
SUM	Sum the values.
PROD	Multiply the values.
MAX	Return the maximum value.
MIN	Return the minimum value.

Definition at line 512 of file CudaDnn.cs.

◆ OP

enum MyCaffe.common.OP

Defines the operations performed by the channel_op function.

Enumerator
MUL	Specifies to perform a multiplication operation.
DIV	Specifies to perform a division operation.
ADD	Specifies to perform an addition operation.
SUB	Specifies to perform a subtraction operation.

Definition at line 134 of file CudaDnn.cs.

◆ ORIENTATION

enum MyCaffe.common.ORIENTATION

Specifies the orientation of a matrix.

See also: CudaDnn::matrix_add_vector

Enumerator
COL	Specifies to add the vector to each column.
ROW	Specifies to add the vector to each row.

Definition at line 644 of file CudaDnn.cs.

◆ POOLING_METHOD

enum MyCaffe.common.POOLING_METHOD

Specifies the pooling method to use when using the Caffe pooling (instead of the pooling from NVIDIA's cuDnn).

See also: CudaDnn::pooling_fwd

Enumerator
MAX	Select the maximum value from the kernel.
AVE	Select the average of the values in the kernel.
STO_TRAIN	Select the stochastic value in the kernel - used during a training pass.
STO_TEST	Select the stochastic value in the kernel - used during a testing pass.

Definition at line 352 of file CudaDnn.cs.

◆ PoolingMethod

enum MyCaffe.common.PoolingMethod

Specifies the pooling method used by the cuDnn function SetPoolingDesc.

See also: CudaDnn::SetPoolingDesc

Enumerator

MAX

Specifies to use

CUDNN_POOLING_MAX in CUDA C++ code.

AVE

Specifies to use

CUDNN_POOLING_AVERAGE_COUNT_INCLUDE_PADDING in CUDA C++ code.

Definition at line 176 of file CudaDnn.cs.

◆ RNN_BIAS_MODE

enum MyCaffe.common.RNN_BIAS_MODE

Specifies the RNN bias mode to use with the Recurrent Layer when using the cuDNN engine.

Enumerator
RNN_NO_BIAS	Specifies to use no bias in the RNN cells.
RNN_SINGLE_INP_BIAS	Specifies to use one bias in the input Gemm of the rnn cell.
RNN_DOUBLE_BIAS	Specifies to use two bias in the input Gemm and recurrent Gemm of the rnn cell (default).
RNN_SINGLE_REC_BIAS	Specifies to use one recurrent bias in the recurrent Gemm of the rnn cell.

Definition at line 400 of file CudaDnn.cs.

◆ RNN_DATALAYOUT

enum MyCaffe.common.RNN_DATALAYOUT

Specifies the RNN data layout of the data input.

Enumerator
RNN_SEQ_MAJOR_UNPACKED	Specifies ordering with sequence major ordering, and padded outer stride from one time-step to the next.
RNN_SEQ_MAJOR_PACKED	Specifies ordering with sequence major ordering, and sequence length sorted and packed.
RNN_BATCH_MAJOR_UNPACKED	Specifies ordering with batch major ordering, padded, outer stride from one batch to the next.

Definition at line 423 of file CudaDnn.cs.

◆ RNN_DIRECTION

enum MyCaffe.common.RNN_DIRECTION

Specifies the RNN directional used.

Enumerator
RNN_UNIDIRECTIONAL	Specifies a single direction RNN (default)
RNN_BIDIRECTIONAL	Specifies a bi-direction RNN where the output is concatinated at each layer.

Definition at line 442 of file CudaDnn.cs.

◆ RNN_FILLER_TYPE

enum MyCaffe.common.RNN_FILLER_TYPE

Defines the filler types used to fill the RNN8 weights.

Enumerator
RNN_CONSTANT_FILLER	Specifies to fill with a constant value.
RNN_XAVIER_FILLER	Specifies to fill with a uniform distribution.
RNN_GAUSSIAN_FILLER	Specifies to fill with a gaussian distribution.

Definition at line 457 of file CudaDnn.cs.

◆ RNN_MODE

enum MyCaffe.common.RNN_MODE

Specifies the RNN mode to use with the Recurrent Layer when using the cuDNN engine.

Enumerator
RNN_RELU	Specifies to use a single RelU gate Recurrent Learning unit.
RNN_TANH	Specifies to use a single TanH gate Recurrent Learning unit.
LSTM	Specifies to use a 4 gate LSTM Recurrent Learning unit.
GRU	Specifies to use the GRU RNN where and

Definition at line 375 of file CudaDnn.cs.

◆ SOFTMAX_ALGORITHM

enum MyCaffe.common.SOFTMAX_ALGORITHM

Specifies the SOFTMAX algorithm to use.

Enumerator
DEFAULT	Specifies to use the default algorithm.
FAST	Specifies to use the fast algorithm.
ACCURATE	Specifies to use the accurate algorithm.
LOG	Specifies to use the log algorithm.

Definition at line 700 of file CudaDnn.cs.

◆ SOFTMAX_MODE

enum MyCaffe.common.SOFTMAX_MODE

Specifies the SOFTMAX mode to use.

Enumerator
INSTANCE	Specifies to run the softmax separately for each N, across CHW dimensions.
CHANNEL	Specifies to run the softmax separately for each N*C, across HW dimensions.

Definition at line 723 of file CudaDnn.cs.

◆ SSD_CODE_TYPE

enum MyCaffe.common.SSD_CODE_TYPE

Defines the encode/decode type used during SSD cuda training.

This enum matches the values of the PriorBoxParameter.CodeType with the values supported in the low level CudaDnnDll.

Enumerator
CORNER	Encode the corner.
CENTER_SIZE	Encode the center size.
CORNER_SIZE	Encode the corner size.

Definition at line 584 of file CudaDnn.cs.

◆ SSD_CONF_LOSS_TYPE

enum MyCaffe.common.SSD_CONF_LOSS_TYPE

Defines the confidence loss types used during SSD cuda training.

This enum matches the values of the MultiboxLossParameter.ConfLossType with the values supported in the low level CudaDnnDll.

Enumerator
SOFTMAX	Specifies to use softmax.
LOGISTIC	Specifies to use logistic.

Definition at line 607 of file CudaDnn.cs.

◆ SSD_LOC_LOSS_TYPE

enum MyCaffe.common.SSD_LOC_LOSS_TYPE

Defines the location loss types used during SSD cuda training.

This enum matches the values of the MultiboxLossParameter.LocLossType with the values supported in the low level CudaDnnDll.

Enumerator
L2	Specifies to use L2 loss.
SMOOTH_L1	Specifies to use smooth L1 loss.

Definition at line 626 of file CudaDnn.cs.

◆ SSD_MATCH_TYPE

enum MyCaffe.common.SSD_MATCH_TYPE

Defines the matching method used during SSD cuda training.

This enum matches the values of the MultiBoxLossParameter.MatchType with the values supported in the low level CudaDnnDll.

Enumerator
BIPARTITE	Specifies to use Bi-Partite.
PER_PREDICTION	Specifies to use per-prediction matching.

Definition at line 565 of file CudaDnn.cs.

◆ SSD_MINING_TYPE

enum MyCaffe.common.SSD_MINING_TYPE

Defines the mining type used during SSD cuda training.

This enum matches the values of the MultiBoxLossParameter.MiningType with the values supported in the low level CudaDnnDll.

Enumerator

NONE

Use all negatives.

MAX_NEGATIVE

Select negatives based on the score.

HARD_EXAMPLE

Select hard examples based on Shrivastava et. al. method.

See also: Training Region-based Object Detectors with Online Hard Example Mining by Abhinav Shrivastava, Abhinav Gupta, Ross Girshick, 2016.

Definition at line 539 of file CudaDnn.cs.

◆ TRAIN_STEP

enum MyCaffe.common.TRAIN_STEP

Defines the training stepping method (if any).

Enumerator
NONE	No stepping.
FORWARD	Step only in the forward direction.
BACKWARD	Step only in the backward direction.
BOTH	Step in both directions (one forward and one backward).

Definition at line 130 of file Interfaces.cs.

◆ TRANSPOSE_OPERATION

enum MyCaffe.common.TRANSPOSE_OPERATION

Specifies the type of operation to perform along with a matrix transposition.

See also: CudaDnn::matrix_transpose_operation

Enumerator
ADD	Add the matrix values after transposing.
MUL	Multiply the matrix values after transposing.
DIV	Divide the matrix values after transposing.

Definition at line 662 of file CudaDnn.cs.

◆ WEIGHT_TARGET

enum MyCaffe.common.WEIGHT_TARGET

Defines the type of weight to target in re-initializations.

Enumerator
NONE	No weights are targeted.
WEIGHTS	Generic weights are targeted.
BIAS	Bias weights are targeted.
BOTH	Both weights and bias are targeted.

Definition at line 37 of file Interfaces.cs.

Function Documentation

◆ onGetWorkspace()

delegate void MyCaffe.common.onGetWorkspace	(	object	sender,
		WorkspaceArgs	e
	)

Delegate used to set the OnGetworkspace event.

Parameters

sender	Specifies the sender.
e	Specifies the arguments.

◆ onSetWorkspace()

delegate void MyCaffe.common.onSetWorkspace	(	object	sender,
		WorkspaceArgs	e
	)

Delegate used to set the OnSetworkspace event.