Class/Object

com.intel.analytics.zoo.models.caffe

CaffeLoader

Related Docs: object CaffeLoader | package caffe

Permalink

class CaffeLoader[T] extends AnyRef

An utility to load pre-trained caffe model from prototxt and binary and convert it to Zoo equivalent modules

T

type

Linear Supertypes
AnyRef, Any
Ordering
  1. Alphabetic
  2. By Inheritance
Inherited
  1. CaffeLoader
  2. AnyRef
  3. Any
  1. Hide All
  2. Show All
Visibility
  1. Public
  2. All

Instance Constructors

  1. new CaffeLoader(prototxtPath: String, modelPath: String, matchAll: Boolean = true, customizedConverters: HashMap[String, Customizable[T]] = null)(implicit arg0: ClassTag[T], ev: TensorNumeric[T])

    Permalink

    prototxtPath

    caffe model define prototxt path

    modelPath

    caffe serialized binary model path

    matchAll

    if match all modules with parameters

    customizedConverters

    customized converter

Value Members

  1. final def !=(arg0: Any): Boolean

    Permalink
    Definition Classes
    AnyRef → Any

  2. final def ##(): Int

    Permalink
    Definition Classes
    AnyRef → Any

  3. final def ==(arg0: Any): Boolean

    Permalink
    Definition Classes
    AnyRef → Any

  4. final def asInstanceOf[T0]: T0

    Permalink
    Definition Classes
    Any

  5. def clone(): AnyRef

    Permalink
    Attributes
    protected[java.lang]
    Definition Classes
    AnyRef
    Annotations
    @throws( ... )

  6. def createCaffeModel(outputNames: Array[String] = Array[String]()): (Module[T], ParallelCriterion[T])

    Permalink

    Load caffe model from prototxt file and binary pre-trained model and converted to Zoo graph module

    Load caffe model from prototxt file and binary pre-trained model and converted to Zoo graph module

    outputNames

    additional output layer names besides the default(layers without next nodes)

    returns

    Zoo model and criterion

  7. final def eq(arg0: AnyRef): Boolean

    Permalink
    Definition Classes
    AnyRef

  8. def equals(arg0: Any): Boolean

    Permalink
    Definition Classes
    AnyRef → Any

  9. def finalize(): Unit

    Permalink
    Attributes
    protected[java.lang]
    Definition Classes
    AnyRef
    Annotations
    @throws( classOf[java.lang.Throwable] )

  10. final def getClass(): Class[_]

    Permalink
    Definition Classes
    AnyRef → Any

  11. def hashCode(): Int

    Permalink
    Definition Classes
    AnyRef → Any

  12. final def isInstanceOf[T0]: Boolean

    Permalink
    Definition Classes
    Any

  13. final def ne(arg0: AnyRef): Boolean

    Permalink
    Definition Classes
    AnyRef

  14. final def notify(): Unit

    Permalink
    Definition Classes
    AnyRef

  15. final def notifyAll(): Unit

    Permalink
    Definition Classes
    AnyRef

  16. def searchAndMergeBnScale(curNode: ModuleNode[T]): Unit

    Permalink

    This method recursively modify the nodes information and merge the BatchNorm layer and Scale layer Specifically, given a node in Graph, it search the nextNodes of this node, and copy the parameter of Scale to BatchNorm, then operate the pointer of BatchNorm to skip Scale layer and directly pointer to the next layer of Scale layer

    This method recursively modify the nodes information and merge the BatchNorm layer and Scale layer Specifically, given a node in Graph, it search the nextNodes of this node, and copy the parameter of Scale to BatchNorm, then operate the pointer of BatchNorm to skip Scale layer and directly pointer to the next layer of Scale layer

    curNode

    the node to be operated recursively

  17. final def synchronized[T0](arg0: ⇒ T0): T0

    Permalink
    Definition Classes
    AnyRef

  18. def toDnnCaffe(graph: StaticGraph[T]): Unit

    Permalink

    Original loading caffe model would bring Scale layer, which is not supported by mkldnn, thus leads to bad performance this method basically change the nodes information of the old graph, nodes information contains all the information of a net, thus only by modifying nodes we can get the new graph Besides, many initializations are based on nodes information, thus, a new construction of Graph is required after nodes are changed

    Original loading caffe model would bring Scale layer, which is not supported by mkldnn, thus leads to bad performance this method basically change the nodes information of the old graph, nodes information contains all the information of a net, thus only by modifying nodes we can get the new graph Besides, many initializations are based on nodes information, thus, a new construction of Graph is required after nodes are changed

    graph

    origin graph which is not supported by mkldnn

  19. def toString(): String

    Permalink
    Definition Classes
    AnyRef → Any

  20. final def wait(): Unit

    Permalink
    Definition Classes
    AnyRef
    Annotations
    @throws( ... )

  21. final def wait(arg0: Long, arg1: Int): Unit

    Permalink
    Definition Classes
    AnyRef
    Annotations
    @throws( ... )

  22. final def wait(arg0: Long): Unit

    Permalink
    Definition Classes
    AnyRef
    Annotations
    @throws( ... )

Inherited from AnyRef

Inherited from Any

Ungrouped