cnn.CNNModelHelper Class Reference

Inheritance diagram for cnn.CNNModelHelper:

Public Member Functions
def	__init__ (self, order="NCHW", name=None, use_cudnn=True, cudnn_exhaustive_search=False, ws_nbytes_limit=None, init_params=True, skip_sparse_optim=False, param_model=None)
def	GetWeights (self, namescope=None)
def	GetBiases (self, namescope=None)
def	ImageInput (self, blob_in, blob_out, use_gpu_transform=False, kwargs)
def	PadImage (self, blob_in, blob_out, kwargs)
def	ConvNd (self, args, kwargs)
def	Conv (self, args, kwargs)
def	ConvTranspose (self, args, kwargs)
def	GroupConv (self, args, kwargs)
def	GroupConv_Deprecated (self, args, kwargs)
def	FC (self, args, kwargs)
def	PackedFC (self, args, kwargs)
def	FC_Prune (self, args, kwargs)
def	FC_Decomp (self, args, kwargs)
def	FC_Sparse (self, args, kwargs)
def	Dropout (self, args, kwargs)
def	LRN (self, args, kwargs)
def	Softmax (self, args, kwargs)
def	SpatialBN (self, args, kwargs)
def	InstanceNorm (self, args, kwargs)
def	Relu (self, args, kwargs)
def	PRelu (self, args, kwargs)
def	Concat (self, args, kwargs)
def	DepthConcat (self, args, kwargs)
def	Sum (self, args, kwargs)
def	Transpose (self, args, kwargs)
def	Iter (self, args, kwargs)
def	Accuracy (self, args, kwargs)
def	MaxPool (self, args, kwargs)
def	AveragePool (self, args, kwargs)
def	XavierInit (self)
def	ConstantInit (self, value)
def	MSRAInit (self)
def	ZeroInit (self)
def	AddWeightDecay (self, weight_decay)
def	CPU (self)
def	GPU (self, gpu_id=0)

Public Attributes
	order
	use_cudnn
	cudnn_exhaustive_search
	ws_nbytes_limit

Detailed Description

A helper model so we can write CNN models more easily, without having to
manually define parameter initializations and operators separately.

Definition at line 13 of file cnn.py.

Member Function Documentation

AddWeightDecay()

def cnn.CNNModelHelper.AddWeightDecay	(		self,
			weight_decay
	)

Adds a decay to weights in the model.

This is a form of L2 regularization.

Args:
    weight_decay: strength of the regularization

Definition at line 200 of file cnn.py.

DepthConcat()

def cnn.CNNModelHelper.DepthConcat	(		self,
			args,
			kwargs
	)

The old depth concat function - we should move to use concat.

Definition at line 159 of file cnn.py.

ImageInput()

def cnn.CNNModelHelper.ImageInput	(		self,
			blob_in,
			blob_out,
			use_gpu_transform = False,
			kwargs
	)

Image Input.

Definition at line 60 of file cnn.py.

---

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

• caffe2/python/cnn.py