These two classes are highlighted as they’re commonly used in examples and tutorials. Workspace is a key component of Caffe2 while CNNModelHelper is useful in quickly creating CNNs.

Workspace

Code: workspace.py | API Docs: Module caffe2.python.workspace

Workspace is a class that holds all the related objects created during runtime:

  1. all blobs, and
  2. all instantiated networks. It is the owner of all these objects and deals with the scaffolding logistics.

Example Usage

A workspace is created for you whenever you create nets or handle blobs of data with Caffe2. Calling workspace initializes an empty workspace with the given root folder. For any operators that are going to interface with the file system, such as load operators, they will write things under this root folder given by the workspace.

1
2
3
4
5
6
7
8
9
10
11
12
from caffe2.proto.caffe2_pb2 import NetDef
from caffe2.python import workspace

init_net = NetDef()
init_net.ParseFromString(open(protobuf_data))
predict_net = NetDef()
predict_net.ParseFromString(open(protobuf_data))
print predict_net.name //would reveal the name field

workspace.CreateNet(init_net)
workspace.CreateNet(predict_net)
workspace.RunNet(predict_net)

You will notice that the CreateNet methods require a net. These are NetDefs that are created from data from datasets, pre-trained models, protobuf data describing the nets and models need to be instantiated as prototype objects inheriting the characteristics from Caffe2’s protobuf spec. These will then be managed in the workspace. Also, note that we’ve loaded the init_net first, which will setup references to blobs that should be filled by predict_net once you load that net.

For more examples of the basics of workspaces, check out the basics tutorial.

CreateNet

Creates an empty net unless blobs are passed in.

Inputs  
net required NetDef
input_blobs  
Outputs
net object 
1
workspace.CreateNet(net_def, input_blobs)

FeedBlob

Feeds a blob into the workspace.

Inputs  
name the name of the blob
arr either a TensorProto object or a numpy array object to be fed into the workspace
device_option (optional) the device option to feed the data with
Returns  
  True or False, stating whether the feed is successful. 
1
workspace.FeedBlob(name, arr, device_option=None)

FetchBlob

Fetches a blob from the workspace.

Inputs  
name the name of the blob - a string or a BlobReference
Returns  
  Fetched blob (numpy array or string) if successful 
1
workspace.FetchBlob(name)

FetchBlobs

Fetches a list of blobs from the workspace.

Inputs  
names list of names of blobs - strings or BlobReferences
Returns  
  list of fetched blobs 
1
workspace.FetchBlob(name)

GetNameScope

Returns the current namescope string. To be used to fetch blobs.

Outputs
namescope

InferShapesAndTypes

Infers the shapes and types for the specified nets.

Inputs  
nets the list of nets
blob_dimensions (optional) a dictionary of blobs and their dimensions. If not specified, the workspace blobs are used.
Returns  
  A tuple of (shapes, types) dictionaries keyed by blob name. 
1
workspace.InferShapesAndTypes(nets, blob_dimensions)

ResetWorkSpace

Resets the workspace, and if root_folder is empty it will keep the current folder setting.

Inputs  
root_folder string
Outputs
workspace object 
1
workspace.ResetWorkspace(root_folder)

RunNet

Runs a given net.

Inputs  
name the name of the net, or a reference to the net.
num_iter number of iterations to run, defaults to 1
Returns  
  True or an exception. 
1
workspace.RunNet(name, num_iter)

RunNetOnce

Takes in a net and will run the net one time.

Inputs
net 
1
workspace.RunNetOnce(net)

RunOperatorOnce

Will execute a single operator.

Inputs
operator 
1
workspace.RunOperatorOnce(operator)

RunOperatorsOnce

Will execute a set of operators.

Inputs
operators list
Outputs
Boolean on success
False if any op fails 
1
workspace.RunOperatorOnce(operators)

RunPlan

Construct a plan of multiple execution steps to run multiple different networks. Use RunPlan to execute this plan.

Inputs
plan_or_step
Outputs
protobuf 
1
workspace.RunPlan(plan_or_step)

StartMint

Starts a mint instance. Note: this does not work well under ipython yet. According to https://github.com/ipython/ipython/issues/5862

Inputs  
root_folder string
port int
Output  
  mint instance 
1
workspace.StartMint(root_folder, port)

StringifyBlobName

Returns the name of a blob.

Inputs
name
Outputs
name, “BlobReference” 
1
workspace.StringifyBlobName(name)

StringifyNetName

Returns the name of a net.

Inputs
name
Outputs
name, “Net” 
1
workspace.StringifyNetName(name)

StringifyProto

Stringify a protocol buffer object.

Inputs  
obj a protocol buffer object, or a Pycaffe2 object that has a Proto() function.
Outputs  
string the output protobuf string.
Raises  
AttributeError if the passed in object does not have the right attribute. 
1
workspace.StringifyProto(name)

Edit on GitHub