Many resources are needed to download a project. Please understand that we have to compensate our server costs. Thank you in advance.
Project price only 1 $
You can buy this project and download/modify it how often you want.
com.intel.analytics.bigdl.models.inception.Inception_v1.scala Maven / Gradle / Ivy
/*
* Copyright 2016 The BigDL Authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package com.intel.analytics.bigdl.models.inception
import com.intel.analytics.bigdl._
import com.intel.analytics.bigdl.nn.{Graph, _}
import com.intel.analytics.bigdl.numeric.NumericFloat
import com.intel.analytics.bigdl.utils.{T, Table}
import com.intel.analytics.bigdl.nn.Graph.ModuleNode
import com.intel.analytics.bigdl.Module
object Inception_Layer_v1 {
def apply(inputSize: Int, config: Table, namePrefix : String = "") : Module[Float] = {
val concat = Concat(2)
val conv1 = Sequential()
conv1.add(SpatialConvolution(inputSize,
config[Table](1)(1), 1, 1, 1, 1)
.setInitMethod(weightInitMethod = Xavier, Zeros).setName(namePrefix + "1x1"))
conv1.add(ReLU(true).setName(namePrefix + "relu_1x1"))
concat.add(conv1)
val conv3 = Sequential()
conv3.add(SpatialConvolution(inputSize,
config[Table](2)(1), 1, 1, 1, 1)
.setInitMethod(weightInitMethod = Xavier, Zeros).setName(namePrefix + "3x3_reduce"))
conv3.add(ReLU(true).setName(namePrefix + "relu_3x3_reduce"))
conv3.add(SpatialConvolution(config[Table](2)(1),
config[Table](2)(2), 3, 3, 1, 1, 1, 1)
.setInitMethod(weightInitMethod = Xavier, Zeros).setName(namePrefix + "3x3"))
conv3.add(ReLU(true).setName(namePrefix + "relu_3x3"))
concat.add(conv3)
val conv5 = Sequential()
conv5.add(SpatialConvolution(inputSize,
config[Table](3)(1), 1, 1, 1, 1)
.setInitMethod(weightInitMethod = Xavier, Zeros).setName(namePrefix + "5x5_reduce"))
conv5.add(ReLU(true).setName(namePrefix + "relu_5x5_reduce"))
conv5.add(SpatialConvolution(config[Table](3)(1),
config[Table](3)(2), 5, 5, 1, 1, 2, 2)
.setInitMethod(weightInitMethod = Xavier, Zeros).setName(namePrefix + "5x5"))
conv5.add(ReLU(true).setName(namePrefix + "relu_5x5"))
concat.add(conv5)
val pool = Sequential()
pool.add(SpatialMaxPooling(3, 3, 1, 1, 1, 1).ceil().setName(namePrefix + "pool"))
pool.add(SpatialConvolution(inputSize,
config[Table](4)(1), 1, 1, 1, 1)
.setInitMethod(weightInitMethod = Xavier, Zeros).setName(namePrefix + "pool_proj"))
pool.add(ReLU(true).setName(namePrefix + "relu_pool_proj"))
concat.add(pool).setName(namePrefix + "output")
concat
}
def apply(input: ModuleNode[Float], inputSize: Int, config: Table, namePrefix : String)
: ModuleNode[Float] = {
val conv1x1 = SpatialConvolution(inputSize, config[Table](1)(1), 1, 1, 1, 1)
.setInitMethod(weightInitMethod = Xavier, Zeros).setName(namePrefix + "1x1").inputs(input)
val relu1x1 = ReLU(true).setName(namePrefix + "relu_1x1").inputs(conv1x1)
val conv3x3_1 = SpatialConvolution(inputSize, config[Table](2)(1), 1, 1, 1, 1).setInitMethod(
weightInitMethod = Xavier, Zeros).setName(namePrefix + "3x3_reduce").inputs(input)
val relu3x3_1 = ReLU(true).setName(namePrefix + "relu_3x3_reduce").inputs(conv3x3_1)
val conv3x3_2 = SpatialConvolution(
config[Table](2)(1), config[Table](2)(2), 3, 3, 1, 1, 1, 1)
.setInitMethod(weightInitMethod = Xavier, Zeros).setName(namePrefix + "3x3").inputs(relu3x3_1)
val relu3x3_2 = ReLU(true).setName(namePrefix + "relu_3x3").inputs(conv3x3_2)
val conv5x5_1 = SpatialConvolution(inputSize, config[Table](3)(1), 1, 1, 1, 1).setInitMethod(
weightInitMethod = Xavier, Zeros).setName(namePrefix + "5x5_reduce").inputs(input)
val relu5x5_1 = ReLU(true).setName(namePrefix + "relu_5x5_reduce").inputs(conv5x5_1)
val conv5x5_2 = SpatialConvolution(
config[Table](3)(1), config[Table](3)(2), 5, 5, 1, 1, 2, 2)
.setInitMethod(weightInitMethod = Xavier, Zeros).setName(namePrefix + "5x5").inputs(relu5x5_1)
val relu5x5_2 = ReLU(true).setName(namePrefix + "relu_5x5").inputs(conv5x5_2)
val pool = SpatialMaxPooling(3, 3, 1, 1, 1, 1).ceil()
.setName(namePrefix + "pool").inputs(input)
val convPool = SpatialConvolution(inputSize, config[Table](4)(1), 1, 1, 1, 1).setInitMethod(
weightInitMethod = Xavier, Zeros).setName(namePrefix + "pool_proj").inputs(pool)
val reluPool = ReLU(true).setName(namePrefix + "relu_pool_proj").inputs(convPool)
JoinTable(2, 0).inputs(relu1x1, relu3x3_2, relu5x5_2, reluPool)
}
}
object Inception_v1_NoAuxClassifier {
def apply(classNum: Int, hasDropout: Boolean = true): Module[Float] = {
val model = Sequential()
model.add(SpatialConvolution(3, 64, 7, 7, 2, 2, 3, 3, 1, false)
.setInitMethod(weightInitMethod = Xavier, Zeros)
.setName("conv1/7x7_s2"))
model.add(ReLU(true).setName("conv1/relu_7x7"))
model.add(SpatialMaxPooling(3, 3, 2, 2).ceil().setName("pool1/3x3_s2"))
model.add(SpatialCrossMapLRN(5, 0.0001, 0.75).setName("pool1/norm1"))
model.add(SpatialConvolution(64, 64, 1, 1, 1, 1).setInitMethod(weightInitMethod = Xavier, Zeros)
.setName("conv2/3x3_reduce"))
model.add(ReLU(true).setName("conv2/relu_3x3_reduce"))
model.add(SpatialConvolution(64, 192, 3, 3, 1, 1, 1, 1)
.setInitMethod(weightInitMethod = Xavier, Zeros).setName("conv2/3x3"))
model.add(ReLU(true).setName("conv2/relu_3x3"))
model.add(SpatialCrossMapLRN(5, 0.0001, 0.75). setName("conv2/norm2"))
model.add(SpatialMaxPooling(3, 3, 2, 2).ceil().setName("pool2/3x3_s2"))
model.add(Inception_Layer_v1(192, T(T(64), T(96, 128), T(16, 32), T(32)), "inception_3a/"))
model.add(Inception_Layer_v1(256, T(T(128), T(128, 192), T(32, 96), T(64)), "inception_3b/"))
model.add(SpatialMaxPooling(3, 3, 2, 2).ceil().setName("pool3/3x3_s2"))
model.add(Inception_Layer_v1(480, T(T(192), T(96, 208), T(16, 48), T(64)), "inception_4a/"))
model.add(Inception_Layer_v1(512, T(T(160), T(112, 224), T(24, 64), T(64)), "inception_4b/"))
model.add(Inception_Layer_v1(512, T(T(128), T(128, 256), T(24, 64), T(64)), "inception_4c/"))
model.add(Inception_Layer_v1(512, T(T(112), T(144, 288), T(32, 64), T(64)), "inception_4d/"))
model.add(Inception_Layer_v1(528, T(T(256), T(160, 320), T(32, 128), T(128)), "inception_4e/"))
model.add(SpatialMaxPooling(3, 3, 2, 2).ceil().setName("pool4/3x3_s2"))
model.add(Inception_Layer_v1(832, T(T(256), T(160, 320), T(32, 128), T(128)), "inception_5a/"))
model.add(Inception_Layer_v1(832, T(T(384), T(192, 384), T(48, 128), T(128)), "inception_5b/"))
model.add(SpatialAveragePooling(7, 7, 1, 1).setName("pool5/7x7_s1"))
if (hasDropout) model.add(Dropout(0.4).setName("pool5/drop_7x7_s1"))
model.add(View(1024).setNumInputDims(3))
model.add(Linear(1024, classNum)
.setInitMethod(weightInitMethod = Xavier, Zeros).setName("loss3/classifier"))
model.add(LogSoftMax().setName("loss3/loss3"))
model
}
def graph(classNum: Int, hasDropout: Boolean = true)
: Module[Float] = {
val input = SpatialConvolution(3, 64, 7, 7, 2, 2, 3, 3, 1, false)
.setInitMethod(weightInitMethod = Xavier, Zeros).setName("conv1/7x7_s2").inputs()
val conv1_relu = ReLU(true).setName("conv1/relu_7x7").inputs(input)
val pool1_s2 = SpatialMaxPooling(3, 3, 2, 2).ceil().setName("pool1/3x3_s2").inputs(conv1_relu)
val pool1_norm1 = SpatialCrossMapLRN(5, 0.0001, 0.75).setName("pool1/norm1").inputs(pool1_s2)
val conv2 = SpatialConvolution(64, 64, 1, 1, 1, 1).setInitMethod(weightInitMethod = Xavier,
Zeros).setName("conv2/3x3_reduce").inputs(pool1_norm1)
val conv2_relu = ReLU(true).setName("conv2/relu_3x3_reduce").inputs(conv2)
val conv2_3x3 = SpatialConvolution(64, 192, 3, 3, 1, 1, 1, 1)
.setInitMethod(weightInitMethod = Xavier, Zeros).setName("conv2/3x3").inputs(conv2_relu)
val conv2_relu_3x3 = ReLU(true).setName("conv2/relu_3x3").inputs(conv2_3x3)
val conv2_norm2 = SpatialCrossMapLRN(5, 0.0001, 0.75)
.setName("conv2/norm2").inputs(conv2_relu_3x3)
val pool2_s2 = SpatialMaxPooling(3, 3, 2, 2).ceil().setName("pool2/3x3_s2").inputs(conv2_norm2)
val inception_3a = Inception_Layer_v1(pool2_s2, 192,
T(T(64), T(96, 128), T(16, 32), T(32)), "inception_3a/")
val inception_3b = Inception_Layer_v1(inception_3a, 256,
T(T(128), T(128, 192), T(32, 96), T(64)), "inception_3b/")
val pool3 = SpatialMaxPooling(3, 3, 2, 2).ceil().setName("pool3/3x3_s2").inputs(inception_3b)
val inception_4a = Inception_Layer_v1(pool3, 480,
T(T(192), T(96, 208), T(16, 48), T(64)), "inception_4a/")
val inception_4b = Inception_Layer_v1(inception_4a, 512,
T(T(160), T(112, 224), T(24, 64), T(64)), "inception_4b/")
val inception_4c = Inception_Layer_v1(inception_4b, 512,
T(T(128), T(128, 256), T(24, 64), T(64)), "inception_4c/")
val inception_4d = Inception_Layer_v1(inception_4c, 512,
T(T(112), T(144, 288), T(32, 64), T(64)), "inception_4d/")
val inception_4e = Inception_Layer_v1(inception_4d, 528,
T(T(256), T(160, 320), T(32, 128), T(128)), "inception_4e/")
val pool4 = SpatialMaxPooling(3, 3, 2, 2).ceil().setName("pool4/3x3_s2").inputs(inception_4e)
val inception_5a = Inception_Layer_v1(pool4, 832,
T(T(256), T(160, 320), T(32, 128), T(128)), "inception_5a/")
val inception_5b = Inception_Layer_v1(inception_5a,
832, T(T(384), T(192, 384), T(48, 128), T(128)), "inception_5b/")
val pool5 = SpatialAveragePooling(7, 7, 1, 1).setName("pool5/7x7_s1").inputs(inception_5b)
val drop = if (hasDropout) Dropout(0.4).setName("pool5/drop_7x7_s1").inputs(pool5) else pool5
val view = View(1024).setNumInputDims(3).inputs(drop)
val classifier = Linear(1024, classNum).setInitMethod(weightInitMethod = Xavier, Zeros)
.setName("loss3/classifier").inputs(view)
val loss = LogSoftMax().setName("loss3/loss3").inputs(classifier)
Graph(input, loss)
}
}
object Inception_v1 {
def apply(classNum: Int, hasDropout: Boolean = true): Module[Float] = {
val feature1 = Sequential()
feature1.add(SpatialConvolution(3, 64, 7, 7, 2, 2, 3, 3, 1, false)
.setInitMethod(weightInitMethod = Xavier, Zeros)
.setName("conv1/7x7_s2"))
feature1.add(ReLU(true).setName("conv1/relu_7x7"))
feature1.add(SpatialMaxPooling(3, 3, 2, 2).ceil().setName("pool1/3x3_s2"))
feature1.add(SpatialCrossMapLRN(5, 0.0001, 0.75).setName("pool1/norm1"))
feature1.add(SpatialConvolution(64, 64, 1, 1, 1, 1)
.setInitMethod(weightInitMethod = Xavier, Zeros)
.setName("conv2/3x3_reduce"))
feature1.add(ReLU(true).setName("conv2/relu_3x3_reduce"))
feature1.add(SpatialConvolution(64, 192, 3, 3, 1, 1, 1, 1)
.setInitMethod(weightInitMethod = Xavier, Zeros)
.setName("conv2/3x3"))
feature1.add(ReLU(true).setName("conv2/relu_3x3"))
feature1.add(SpatialCrossMapLRN(5, 0.0001, 0.75). setName("conv2/norm2"))
feature1.add(SpatialMaxPooling(3, 3, 2, 2).ceil().setName("pool2/3x3_s2"))
feature1.add(Inception_Layer_v1(192, T(T(64), T(96, 128), T(16, 32), T(32)), "inception_3a/"))
feature1.add(Inception_Layer_v1(256, T(T(128), T(128, 192), T(32, 96), T(64)), "inception_3b/"))
feature1.add(SpatialMaxPooling(3, 3, 2, 2).ceil().setName("pool3/3x3_s2"))
feature1.add(Inception_Layer_v1(480, T(T(192), T(96, 208), T(16, 48), T(64)), "inception_4a/"))
val output1 = Sequential()
output1.add(SpatialAveragePooling(5, 5, 3, 3).ceil().setName("loss1/ave_pool"))
output1.add(SpatialConvolution(512, 128, 1, 1, 1, 1).setName("loss1/conv"))
output1.add(ReLU(true).setName("loss1/relu_conv"))
output1.add(View(128 * 4 * 4).setNumInputDims(3))
output1.add(Linear(128 * 4 * 4, 1024).setName("loss1/fc"))
output1.add(ReLU(true).setName("loss1/relu_fc"))
if (hasDropout) output1.add(Dropout(0.7).setName("loss1/drop_fc"))
output1.add(Linear(1024, classNum).setName("loss1/classifier"))
output1.add(LogSoftMax().setName("loss1/loss"))
val feature2 = Sequential()
feature2.add(Inception_Layer_v1(512, T(T(160), T(112, 224), T(24, 64), T(64)), "inception_4b/"))
feature2.add(Inception_Layer_v1(512, T(T(128), T(128, 256), T(24, 64), T(64)), "inception_4c/"))
feature2.add(Inception_Layer_v1(512, T(T(112), T(144, 288), T(32, 64), T(64)), "inception_4d/"))
val output2 = Sequential()
output2.add(SpatialAveragePooling(5, 5, 3, 3).setName("loss2/ave_pool"))
output2.add(SpatialConvolution(528, 128, 1, 1, 1, 1).setName("loss2/conv"))
output2.add(ReLU(true).setName("loss2/relu_conv"))
output2.add(View(128 * 4 * 4).setNumInputDims(3))
output2.add(Linear(128 * 4 * 4, 1024).setName("loss2/fc"))
output2.add(ReLU(true).setName("loss2/relu_fc"))
if (hasDropout) output2.add(Dropout(0.7).setName("loss2/drop_fc"))
output2.add(Linear(1024, classNum).setName("loss2/classifier"))
output2.add(LogSoftMax().setName("loss2/loss"))
val output3 = Sequential()
output3.add(Inception_Layer_v1(528, T(T(256), T(160, 320), T(32, 128), T(128)),
"inception_4e/"))
output3.add(SpatialMaxPooling(3, 3, 2, 2).ceil().setName("pool4/3x3_s2"))
output3.add(Inception_Layer_v1(832, T(T(256), T(160, 320), T(32, 128), T(128)),
"inception_5a/"))
output3.add(Inception_Layer_v1(832, T(T(384), T(192, 384), T(48, 128), T(128)),
"inception_5b/"))
output3.add(SpatialAveragePooling(7, 7, 1, 1).setName("pool5/7x7_s1"))
if (hasDropout) output3.add(Dropout(0.4).setName("pool5/drop_7x7_s1"))
output3.add(View(1024).setNumInputDims(3))
output3.add(Linear(1024, classNum)
.setInitMethod(weightInitMethod = Xavier, Zeros).setName("loss3/classifier"))
output3.add(LogSoftMax().setName("loss3/loss3"))
val split2 = Concat(2).setName("split2")
split2.add(output3)
split2.add(output2)
val mainBranch = Sequential()
mainBranch.add(feature2)
mainBranch.add(split2)
val split1 = Concat(2).setName("split1")
split1.add(mainBranch)
split1.add(output1)
val model = Sequential()
model.add(feature1)
model.add(split1)
model
}
def graph(classNum: Int, hasDropout: Boolean = true): Module[Float] = {
val input = Input()
val conv1 = SpatialConvolution(3, 64, 7, 7, 2, 2, 3, 3, 1, false)
.setInitMethod(weightInitMethod = Xavier, Zeros)
.setName("conv1/7x7_s2").inputs(input)
val relu1 = ReLU(true).setName("conv1/relu_7x7").inputs(conv1)
val pool1 = SpatialMaxPooling(3, 3, 2, 2).ceil().setName("pool1/3x3_s2").inputs(relu1)
val lrn1 = SpatialCrossMapLRN(5, 0.0001, 0.75).setName("pool1/norm1").inputs(pool1)
val conv2 = SpatialConvolution(64, 64, 1, 1, 1, 1)
.setInitMethod(weightInitMethod = Xavier, Zeros)
.setName("conv2/3x3_reduce").inputs(lrn1)
val relu2 = ReLU(true).setName("conv2/relu_3x3_reduce").inputs(conv2)
val conv3 = SpatialConvolution(64, 192, 3, 3, 1, 1, 1, 1)
.setInitMethod(weightInitMethod = Xavier, Zeros)
.setName("conv2/3x3").inputs(relu2)
val relu3 = ReLU(true).setName("conv2/relu_3x3").inputs(conv3)
val lrn2 = SpatialCrossMapLRN(5, 0.0001, 0.75). setName("conv2/norm2").inputs(relu3)
val pool2 = SpatialMaxPooling(3, 3, 2, 2).ceil().setName("pool2/3x3_s2").inputs(lrn2)
val layer1 = Inception_Layer_v1(pool2, 192, T(T(64), T(96, 128), T(16, 32), T(32)),
"inception_3a/")
val layer2 = Inception_Layer_v1(layer1, 256, T(T(128), T(128, 192), T(32, 96), T(64)),
"inception_3b/")
val pool3 = SpatialMaxPooling(3, 3, 2, 2).ceil().setName("pool3/3x3_s2").inputs(layer2)
val feature1 = Inception_Layer_v1(pool3, 480, T(T(192), T(96, 208), T(16, 48), T(64)),
"inception_4a/")
val pool2_1 = SpatialAveragePooling(5, 5, 3, 3).ceil()
.setName("loss1/ave_pool").inputs(feature1)
val loss2_1 = SpatialConvolution(512, 128, 1, 1, 1, 1).setName("loss1/conv").inputs(pool2_1)
val relu2_1 = ReLU(true).setName("loss1/relu_conv").inputs(loss2_1)
val view2_1 = View(128 * 4 * 4).setNumInputDims(3).inputs(relu2_1)
val linear2_1 = Linear(128 * 4 * 4, 1024).setName("loss1/fc").inputs(view2_1)
val relu2_2 = ReLU(true).setName("loss1/relu_fc").inputs(linear2_1)
val drop2_1 = if (hasDropout) Dropout(0.7).setName("loss1/drop_fc").inputs(relu2_2) else relu2_2
val classifier2_1 = Linear(1024, classNum).setName("loss1/classifier").inputs(drop2_1)
val output1 = LogSoftMax().setName("loss1/loss").inputs(classifier2_1)
val layer3_1 = Inception_Layer_v1(feature1, 512, T(T(160), T(112, 224), T(24, 64), T(64)),
"inception_4b/")
val layer3_2 = Inception_Layer_v1(layer3_1, 512, T(T(128), T(128, 256), T(24, 64), T(64)),
"inception_4c/")
val feature2 = Inception_Layer_v1(layer3_2, 512, T(T(112), T(144, 288), T(32, 64), T(64)),
"inception_4d/")
val pool4_1 = SpatialAveragePooling(5, 5, 3, 3).setName("loss2/ave_pool").inputs(feature2)
val conv4_1 = SpatialConvolution(528, 128, 1, 1, 1, 1).setName("loss2/conv").inputs(pool4_1)
val relu4_1 = ReLU(true).setName("loss2/relu_conv").inputs(conv4_1)
val view4_1 = View(128 * 4 * 4).setNumInputDims(3).inputs(relu4_1)
val linear4_1 = Linear(128 * 4 * 4, 1024).setName("loss2/fc").inputs(view4_1)
val relu4_2 = ReLU(true).setName("loss2/relu_fc").inputs(linear4_1)
val drop4_1 = if (hasDropout) Dropout(0.7).setName("loss2/drop_fc").inputs(relu4_2) else relu4_2
val linear4_2 = Linear(1024, classNum).setName("loss2/classifier").inputs(drop4_1)
val output2 = LogSoftMax().setName("loss2/loss").inputs(linear4_2)
val layer5_1 = Inception_Layer_v1(feature2, 528, T(T(256), T(160, 320), T(32, 128), T(128)),
"inception_4e/")
val pool5_1 = SpatialMaxPooling(3, 3, 2, 2).ceil().setName("pool4/3x3_s2").inputs(layer5_1)
val layer5_2 = Inception_Layer_v1(pool5_1, 832, T(T(256), T(160, 320), T(32, 128), T(128)),
"inception_5a/")
val layer5_3 = Inception_Layer_v1(layer5_2, 832, T(T(384), T(192, 384), T(48, 128), T(128)),
"inception_5b/")
val pool5_4 = SpatialAveragePooling(7, 7, 1, 1).setName("pool5/7x7_s1").inputs(layer5_3)
val drop5_1 = if (hasDropout) Dropout(0.4).setName("pool5/drop_7x7_s1")
.inputs(pool5_4) else pool5_4
val view5_1 = View(1024).setNumInputDims(3).inputs(drop5_1)
val linear5_1 = Linear(1024, classNum)
.setInitMethod(weightInitMethod = Xavier, Zeros).setName("loss3/classifier").inputs(view5_1)
val output3 = LogSoftMax().setName("loss3/loss3").inputs(linear5_1)
val split2 = JoinTable(2, 0).setName("split2").inputs(output3, output2, output1)
Graph(input, split2)
}
}
