diff --git a/paddle/gserver/layers/ConvTransBaseLayer.cpp b/paddle/gserver/layers/ConvTransBaseLayer.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..68fb48a38b5187ea81055351c9cf804b01e0f2b4
--- /dev/null
+++ b/paddle/gserver/layers/ConvTransBaseLayer.cpp
@@ -0,0 +1,77 @@
+/* Copyright (c) 2016 Baidu, Inc. All Rights Reserve.
+
+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. */
+
+
+#include "paddle/utils/Logging.h"
+#include "ConvTransBaseLayer.h"
+namespace paddle {
+
+bool ConvTransBaseLayer::init(const LayerMap& layerMap,
+                         const ParameterMap& parameterMap) {
+  /* Initialize the basic parent class */
+  Layer::init(layerMap, parameterMap);
+
+  /* Initialize the convolutional layer parameter */
+  channel_ = config_.num_filters();
+  sharedBiases_ = config_.shared_biases();
+  for (auto& inputConfig : config_.inputs()) {
+    const ConvConfig& conf = inputConfig.conv_conf();
+    padding_.push_back(conf.padding());
+    stride_.push_back(conf.stride());
+    filterSize_.push_back(conf.filter_size());
+    paddingY_.push_back(conf.padding_y());
+    strideY_.push_back(conf.stride_y());
+    filterSizeY_.push_back(conf.filter_size_y());
+    filterPixels_.push_back(filterSize_.back() * filterSizeY_.back());
+    numFilters_.push_back(conf.channels());
+    imgSize_.push_back(conf.img_size());
+    imgPixels_.push_back(imgSize_.back() * imgSize_.back());
+    groups_.push_back(conf.groups());
+    filterChannels_.push_back(conf.filter_channels());
+    outputX_.push_back(conf.output_x());
+    outputs_.push_back(outputX_.back() * outputX_.back());
+  }
+
+  /* initialize the weightList */
+  CHECK(inputLayers_.size() == parameters_.size());
+  for (size_t i = 0; i < inputLayers_.size(); i++) {
+    size_t height, width;
+    height = filterPixels_[i] * filterChannels_[i];
+    width = numFilters_[i];
+
+    // create a new weight
+    CHECK_EQ(parameters_[i]->getSize(), width * height);
+    Weight* w = new Weight(height, width, parameters_[i]);
+    weights_.emplace_back(w);
+  }
+
+  /* initialize the biases_ */
+  if (biasParameter_.get() != NULL) {
+    if (sharedBiases_) {
+      CHECK_EQ((size_t)channel_, biasParameter_->getSize());
+      biases_ =
+          std::unique_ptr<Weight>(new Weight(channel_, 1, biasParameter_));
+    } else {
+      biases_ =
+          std::unique_ptr<Weight>(new Weight(getSize(), 1, biasParameter_));
+    }
+  }
+
+  // default caffe model
+  caffeMode_ = true;
+
+  return true;
+}
+
+}  // namespace paddle
diff --git a/paddle/gserver/layers/ConvTransBaseLayer.h b/paddle/gserver/layers/ConvTransBaseLayer.h
new file mode 100644
index 0000000000000000000000000000000000000000..467a2605697593233035091a5f851a4f7bbab652
--- /dev/null
+++ b/paddle/gserver/layers/ConvTransBaseLayer.h
@@ -0,0 +1,112 @@
+/* Copyright (c) 2016 Baidu, Inc. All Rights Reserve.
+
+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. */
+
+
+#pragma once
+
+#include "Layer.h"
+namespace paddle {
+
+/**
+ * @brief A Base Convolution Layer, which convolves the input image
+ * with learned filters and (optionally) adds biases.
+ */
+
+class ConvTransBaseLayer : public Layer {
+protected:
+  typedef std::vector<int> IntV;
+
+  /// The number of channel in image (the output of the deconv layer).
+  int channel_;
+  /// The x dimension of the padding.
+  IntV padding_;
+  /// The y dimension of the padding.
+  IntV paddingY_;
+  /// The x dimension of the stride.
+  IntV stride_;
+  /// The y dimension of the stride.
+  IntV strideY_;
+  /// The x dimension of a filter kernel.
+  IntV filterSize_;
+  /// The y dimension of a filter kernel.
+  IntV filterSizeY_;
+  /// The number of filters(i.e. the number channels of the deconv layer input)
+  IntV numFilters_;
+  /// The spatial dimensions of input feature map.
+  IntV imgSize_;
+  /// The total pixel size of input feature map.
+  /// imgPixels_ = imgSizeX_ * imgSizeY_.
+  IntV imgPixels_;
+  /// filterPixels_ = filterSizeX_ * filterSizeY_.
+  IntV filterPixels_;
+  /// filterChannels_ = channels_/groups_.
+  IntV filterChannels_;
+  /// The spatial dimensions of output feature map.
+  IntV outputX_;
+  /// The spatial dimensions of output feature map.
+  IntV outputs_;
+  /// Group size, refer to grouped convolution in
+  /// Alex Krizhevsky's paper: when group=2, the first half of the
+  /// filters are only connected to the first half of the input channels,
+  /// and the second half only connected to the second half.
+  IntV groups_;
+  /// Whether the bias is shared for feature in each channel.
+  bool sharedBiases_;
+
+  /// shape of weight: (numChannels * filterPixels_, numFilters)
+  WeightList weights_;
+  /// If shared_biases is false shape of bias: (numFilters_, 1)
+  /// If shared_biases is ture shape of bias:
+  /// (numFilters_ * outputX * outputY, 1)
+  std::unique_ptr<Weight> biases_;
+
+  /// True by default. The only difference is the calculation
+  /// of output size.
+  bool caffeMode_;
+
+public:
+  explicit ConvTransBaseLayer(const LayerConfig& config) : Layer(config) {}
+
+  virtual bool init(const LayerMap& layerMap, const ParameterMap& parameterMap);
+
+  Weight& getWeight(int idx) { return *weights_[idx]; }
+
+  /**
+   * Calculate image size based on caffeMode_ from outputSize.
+   * - input(+padding): 0123456789
+   * - imageSize(+padding) = 10;
+   * - filterSize = 3;
+   * - stride = 2;
+   * - caffeMode_ is true:
+       - output: (012), (234), (456), (678)
+       - outputSize = 4;
+   * - caffeMode_ is false:
+   *   - output: (012), (234), (456), (678), (9)
+   *   - outputSize = 5;
+   */
+
+  int imageSize(int outputSize, int filterSize, int padding, int stride) {
+    int imageSize;
+    if (!caffeMode_) {
+     imageSize =
+         (outputSize - 1) * stride + filterSize - 2 * padding - stride + 1;
+    } else {
+     imageSize = (outputSize - 1) * stride + filterSize - 2 * padding;
+    }
+    CHECK_GE(imageSize, 1);
+    return imageSize;
+  }
+};
+
+}  // namespace paddle
diff --git a/paddle/gserver/layers/ExpandConvTransLayer.cpp b/paddle/gserver/layers/ExpandConvTransLayer.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..56cc042653b60316b4b8ce740d4904ca2e10c8ae
--- /dev/null
+++ b/paddle/gserver/layers/ExpandConvTransLayer.cpp
@@ -0,0 +1,332 @@
+/* Copyright (c) 2016 Baidu, Inc. All Rights Reserve.
+
+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. */
+
+
+#include "paddle/utils/Logging.h"
+#include "paddle/utils/Stat.h"
+#include "ExpandConvTransLayer.h"
+
+namespace paddle {
+
+REGISTER_LAYER(exconvt, ExpandConvTransLayer);
+
+bool ExpandConvTransLayer::init(const LayerMap &layerMap,
+                           const ParameterMap &parameterMap) {
+  /* Initialize the basic convolutional parent class */
+  ConvTransBaseLayer::init(layerMap, parameterMap);
+
+  /* Initialize the projection */
+  for (auto &inputConfig : config_.inputs()) {
+    const ConvConfig &conf = inputConfig.conv_conf();
+    subM_.push_back(conf.channels() / conf.groups());
+    subN_.push_back(conf.output_x() * conf.output_x());
+    subK_.push_back(channel_ * conf.filter_size() * conf.filter_size() /
+                    conf.groups());
+    /* Consistent caffe mode for multiple input */
+    caffeMode_ = conf.caffe_mode();
+  }
+
+  return true;
+}
+
+// Why this is necessary after calling init?
+size_t ExpandConvTransLayer::getSize() {
+  CHECK_NE(inputLayers_.size(), 0UL);
+  imgSizeH_.clear();
+  imgSizeW_.clear();
+  outputH_.clear();
+  outputW_.clear();
+  subN_.clear();
+  size_t layerSize = 0;
+  for (size_t i = 0; i < inputLayers_.size(); i++) {
+    outputH_.push_back(inputLayers_[i]->getOutput().getFrameHeight());
+    outputW_.push_back(inputLayers_[i]->getOutput().getFrameWidth());
+    if (outputH_[i] == 0) outputH_[i] = outputX_[i];
+    if (outputW_[i] == 0) outputW_[i] = outputX_[i];
+    imgSizeH_.push_back(
+        imageSize(outputH_[i], filterSize_[i], padding_[i], stride_[i]));
+    imgSizeW_.push_back(
+        imageSize(outputW_[i], filterSize_[i], padding_[i], stride_[i]));
+    subN_.push_back(outputH_[i] * outputW_[i]);
+    CHECK(layerSize == 0 ||
+            imgSizeH_[i] * imgSizeW_[i] * (size_t)channel_ == layerSize);
+    layerSize = imgSizeH_[i] * imgSizeW_[i] * channel_;
+  }
+  getOutput().setFrameHeight(imgSizeH_[0]);
+  getOutput().setFrameWidth(imgSizeW_[0]);
+  return layerSize;
+}
+
+void ExpandConvTransLayer::resetExpandInput(size_t height, size_t width) {
+  Matrix::resizeOrCreate(expandInput_, height, width, false, useGpu_);
+}
+
+/*void ExpandConvTransLayer::resetConvOutput(size_t batchSize, int inIdx) {
+  Matrix::resizeOrCreate(transOutValue_, batchSize * numFilters_, subN_[inIdx],
+                         false, useGpu_);
+}*/
+
+
+void ExpandConvTransLayer::addSharedBias() {
+  size_t mapW = getSize() / channel_;
+  size_t mapH = getOutputValue()->getElementCnt() / mapW;
+  MatrixPtr out =
+      Matrix::create(getOutputValue()->getData(), mapH, mapW, false, useGpu_);
+
+  Matrix::resizeOrCreate(transOutValue_, mapW, mapH, false, useGpu_);
+
+  out->transpose(transOutValue_, false);  // false means no memory allocation
+  transOutValue_->reshape(transOutValue_->getElementCnt() / channel_,
+                          channel_);
+
+  MatrixPtr bias =
+      Matrix::create(biases_->getW()->getData(), 1,
+                     biases_->getW()->getElementCnt(), false, useGpu_);
+  transOutValue_->addBias(*bias, 1.0f);
+
+  transOutValue_->reshape(mapW, mapH);
+  transOutValue_->transpose(out, false);  // false means no memory allocation
+
+  out->clear();
+  bias->clear();
+}
+
+void ExpandConvTransLayer::addUnsharedBias() {
+  MatrixPtr outValue = getOutputValue();
+  MatrixPtr bias =
+      Matrix::create(biases_->getW()->getData(), 1,
+                     biases_->getW()->getElementCnt(), false, useGpu_);
+  outValue->addBias(*bias, 1.0f);
+}
+
+
+void ExpandConvTransLayer::expandOneFrame(MatrixPtr image, size_t startIdx,
+                                     int inIdx) {
+  resetExpandInput(subK_[inIdx] * groups_[inIdx], subN_[inIdx]);
+  real *imgData = image->getData() + startIdx * image->getWidth();
+  MatrixPtr imageTmp = Matrix::create(
+      imgData, 1, imgSizeH_[inIdx] * imgSizeW_[inIdx] * channel_, false,
+      useGpu_);
+  expandInput_->convExpand(*imageTmp, imgSizeH_[inIdx], imgSizeW_[inIdx],
+                           channel_, filterSize_[inIdx],
+                           filterSize_[inIdx], stride_[inIdx], stride_[inIdx],
+                           padding_[inIdx], padding_[inIdx],
+                           outputH_[inIdx], outputW_[inIdx]);
+  imageTmp->clear();
+}
+
+void ExpandConvTransLayer::expandBackOnce(MatrixPtr imageGrad, int inIdx,
+                                        int startIdx) {
+  int subM = subM_[inIdx];
+  int subN = subN_[inIdx];
+  int subK = subK_[inIdx];
+
+  LayerPtr prevLayer = getPrev(inIdx);
+  if (NULL == prevLayer->getOutputGrad()) {
+    return;
+  }
+
+  expandOneFrame(imageGrad, startIdx, inIdx);
+
+  real *outGradData =
+      prevLayer -> getOutputGrad()->getData()
+                  + startIdx * subN * numFilters_[inIdx];
+
+  real *wgtData = weights_[inIdx]->getW()->getData();
+  real *expInData = expandInput_->getData();
+  for (int g = 0; g < groups_[inIdx]; ++g) {
+    MatrixPtr A =
+        Matrix::create(wgtData, subK, subM, true, useGpu_);  // mark transpose
+    MatrixPtr B = Matrix::create(expInData, subK, subN, false, useGpu_);
+    MatrixPtr C = Matrix::create(outGradData, subM, subN, false, useGpu_);
+    C->mul(A, B, 1, 1);
+
+    A->clear();
+    B->clear();
+    C->clear();
+    wgtData += subK * subM;
+    expInData += subK * subN;
+    outGradData += subM * subN;
+  }
+}
+
+void ExpandConvTransLayer::forward(PassType passType) {
+  Layer::forward(passType);
+
+  /* malloc memory for the output_ if necessary */
+  /* note: one sample correspond to one colum, and the
+   *   transOutValue correspond sample to one row */
+  int batchSize = inputLayers_[0]->getOutputValue()->getHeight();
+  resetOutput(batchSize, getSize());
+
+  MatrixPtr output = nullptr;
+  for (size_t i = 0; i != inputLayers_.size(); ++i) {
+    LayerPtr prevLayer = getPrev(i);
+    output = prevLayer->getOutputValue();
+    REGISTER_TIMER_INFO("shrinkFwd", getName().c_str());
+    shrinkFwd(output, i);
+  }
+
+  /* add the bias-vector */
+  if (biases_.get() != NULL) {
+    if (sharedBiases_) {
+      addSharedBias();
+    } else {
+      addUnsharedBias();
+    }
+  }
+
+  /* activation */
+  forwardActivation();
+}
+
+void ExpandConvTransLayer::shrinkFwd(MatrixPtr output, int inpIdx) {
+  int subM = subM_[inpIdx];
+  int subN = subN_[inpIdx];
+  int subK = subK_[inpIdx];
+
+  size_t batchSize = output->getHeight();
+  MatrixPtr image = getOutputValue();
+
+  /* reset the expand-grad memory */
+  resetExpandInput(subK * groups_[inpIdx], subN);
+
+  real *localData = output->getData();
+  real *imageData = image->getData();
+  for (size_t n = 0; n < batchSize; n++) {
+    real *wgtData = weights_[inpIdx]->getW()->getData();
+    real *expandInData = expandInput_->getData();
+
+    for (int g = 0; g < groups_[inpIdx]; g++) {
+      // create temporary matrix
+      MatrixPtr C = Matrix::create(expandInData, subK, subN, false, useGpu_);
+      MatrixPtr B = Matrix::create(localData, subM, subN, false, useGpu_);
+      MatrixPtr A = Matrix::create(wgtData, subK, subM, false, useGpu_);
+      C->mul(A, B);  // mul
+
+      // clear the temporary matrix
+      A->clear();
+      B->clear();
+      C->clear();
+
+      expandInData += subK * subN;
+      localData += subM * subN;
+      wgtData += subK * subM;
+    }
+
+    // shrink one frame outGrad
+    MatrixPtr oneTmp = Matrix::create(
+        expandInput_->getData(), subK * groups_[inpIdx], subN, false, useGpu_);
+    MatrixPtr vTmp = Matrix::create(
+        imageData, 1,
+        imgSizeH_[inpIdx] * imgSizeW_[inpIdx] * channel_, false,
+        useGpu_);
+    vTmp->convShrink(*oneTmp, imgSizeH_[inpIdx], imgSizeW_[inpIdx],
+                     channel_, filterSize_[inpIdx],
+                     filterSize_[inpIdx], stride_[inpIdx], stride_[inpIdx],
+                     padding_[inpIdx], padding_[inpIdx],
+                     outputH_[inpIdx], outputW_[inpIdx], 1.0f, 1.0f);
+    vTmp->clear();
+    oneTmp->clear();
+
+    // move the data-pointer
+    imageData += imgSizeH_[inpIdx] * imgSizeW_[inpIdx] * channel_;
+  }
+}
+
+void ExpandConvTransLayer::bpropSharedBias(MatrixPtr biases, MatrixPtr v) {
+  size_t mapW = getSize() / channel_;
+  size_t mapH = v->getElementCnt() / mapW;
+  MatrixPtr vTmp = Matrix::create(v->getData(), mapH, mapW, false, useGpu_);
+
+  Matrix::resizeOrCreate(transOutValue_, mapW, mapH, false, useGpu_);
+
+  vTmp->transpose(transOutValue_, false);  // false means no memory allocation
+  vTmp->reshape(transOutValue_->getElementCnt() / channel_, channel_);
+  biases->collectBias(*vTmp, 1.0f);
+}
+
+void ExpandConvTransLayer::bpropBiases(MatrixPtr v) {
+  MatrixPtr biases =
+      Matrix::create(biases_->getWGrad()->getData(), 1,
+                     biases_->getWGrad()->getElementCnt(), false, useGpu_);
+  if (sharedBiases_) {
+    bpropSharedBias(biases, v);
+  } else {
+    biases->collectBias(*v, 1.0f);
+  }
+  biases->clear();
+}
+
+void ExpandConvTransLayer::backward(const UpdateCallback &callback) {
+  backwardActivation();
+
+  MatrixPtr imageGrad = getOutputGrad();
+  if (biases_ && biases_->getWGrad()) {
+    bpropBiases(imageGrad);
+    /* Increasing the number of gradient */
+    biases_->getParameterPtr()->incUpdate(callback);
+  }
+
+  for (size_t i = 0; i != inputLayers_.size(); ++i) {
+    /* First, calculate the input layers error */
+    for (size_t off = 0; off < imageGrad->getHeight(); off++) {
+        expandBackOnce(imageGrad, i, off);
+    }
+    if (weights_[i]->getWGrad()) {
+      /* Then, calculate the W-gradient for the current layer */
+      bpropWeights(imageGrad, i);
+      /* Increasing the number of gradient */
+      weights_[i]->getParameterPtr()->incUpdate(callback);
+    }
+  }
+}
+
+void ExpandConvTransLayer::bpropWeights(MatrixPtr v, int inpIdx) {
+  MatrixPtr weightGrad = weights_[inpIdx]->getWGrad();
+  MatrixPtr outputV = getPrev(inpIdx)->getOutputValue();
+
+  int subM = subM_[inpIdx];
+  int subN = subN_[inpIdx];
+  int subK = subK_[inpIdx];
+  size_t batchSize = outputV->getHeight();
+  resetExpandInput(subK * groups_[inpIdx], subN);
+
+  real *outputData = outputV -> getData();
+
+  for (size_t n = 0; n < batchSize; n++) {  // frame by frame
+    // expand
+    expandOneFrame(v, n, inpIdx);
+    real *wGradData = weightGrad->getData();
+    real *expandInData = expandInput_->getData();
+
+    // expand-mul one-group by one
+    for (int g = 0; g < groups_[inpIdx]; g++) {
+      MatrixPtr A = Matrix::create(expandInData, subK, subN, false, useGpu_);
+      MatrixPtr B = Matrix::create(outputData, subM, subN, true, useGpu_);
+      MatrixPtr C = Matrix::create(wGradData, subK, subM, false, useGpu_);
+      C->mul(A, B, 1, 1);
+
+      A->clear();
+      B->clear();
+      C->clear();
+      outputData += subM * subN;
+      wGradData += subK * subM;
+      expandInData += subK * subN;
+    }
+  }
+}
+
+
+}  // namespace paddle
diff --git a/paddle/gserver/layers/ExpandConvTransLayer.h b/paddle/gserver/layers/ExpandConvTransLayer.h
new file mode 100644
index 0000000000000000000000000000000000000000..f19aa81f3d3c23d2e24f1c49710cb124dd714a54
--- /dev/null
+++ b/paddle/gserver/layers/ExpandConvTransLayer.h
@@ -0,0 +1,106 @@
+/* Copyright (c) 2016 Baidu, Inc. All Rights Reserve.
+
+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. */
+
+
+#pragma once
+
+#include "ConvTransBaseLayer.h"
+#include "paddle/math/Matrix.h"
+#include <vector>
+
+namespace paddle {
+
+/**
+ * @brief A subclass of convolution layer.
+ * This layer expands input and use matrix multiplication to
+ * calculate convolution operation.
+ *
+ * The config file api is img_conv_layer.
+ */
+class ExpandConvTransLayer : public ConvTransBaseLayer {
+protected:
+  /// For expand convolution.
+  /// subM_ = numFilters_ / groups_.
+  IntV subM_;
+  /// subN_ = outputH_ * outputW_.
+  IntV subN_;
+  /// subK_ = channels_ * filterPixels_ * groups_.
+  IntV subK_;
+  /// The spatial dimensions of height of input feature map.
+  IntV imgSizeH_;
+  /// The spatial dimensions of width of input feature map.
+  IntV imgSizeW_;
+  /// The spatial dimensions of height of output feature map.
+  IntV outputH_;
+  /// The spatial dimensions of width of output feature map.
+  IntV outputW_;
+  /// Expand one sample at a time. shape:
+  /// (numChannels * filterPixels_, outputSizeH * outputSizeW)
+  MatrixPtr expandInput_;
+  /// The transpose of output, which is an auxiliary matrix.
+  MatrixPtr transOutValue_;
+
+public:
+  explicit ExpandConvTransLayer(const LayerConfig& config) :
+      ConvTransBaseLayer(config) {}
+
+  ~ExpandConvTransLayer() {}
+
+  bool init(const LayerMap& layerMap, const ParameterMap& parameterMap);
+
+  size_t getSize();
+
+  /**
+   * Create or resize expandInput_.
+   */
+  void resetExpandInput(size_t height, size_t width);
+
+  /**
+   * Create or resize transOutValue_.
+   */
+  void resetConvOutput(size_t batchSize, int inIdx);
+
+  /**
+   * Expand one input sample.
+   */
+  void expandOneFrame(MatrixPtr image, size_t startIdx, int inIdx);
+
+  /**
+   * Expand one output image and perform matrix multiplication.
+   */
+  void expandBackOnce(MatrixPtr image, int inIdx, int startIdx);
+
+  /**
+   * Perform matrix multiplication on one output and then shrink.
+   */
+  void shrinkFwd(MatrixPtr output, int inpIdx);
+
+  /**
+   * Add shared bias.
+   */
+  void addSharedBias();
+
+  /**
+   * Add unshared bias.
+   */
+  void addUnsharedBias();
+  void forward(PassType passType);
+  void bpropSharedBias(MatrixPtr biases, MatrixPtr v);
+  void bpropBiases(MatrixPtr v);
+  void backward(const UpdateCallback& callback);
+  void bpropWeights(MatrixPtr v, int inpIdx);
+  void bpropActs(MatrixPtr v, int inpIdx);
+};
+
+}  // namespace paddle
diff --git a/paddle/gserver/tests/test_LayerGrad.cpp b/paddle/gserver/tests/test_LayerGrad.cpp
index 4e01fa91ed2ba6b40882d9995e52e9dbeb37f57e..d634d198c3be7578c0c330466c57dd76714f55fd 100644
--- a/paddle/gserver/tests/test_LayerGrad.cpp
+++ b/paddle/gserver/tests/test_LayerGrad.cpp
@@ -312,6 +312,49 @@ TEST(Layer, convLayer) {
 #endif
 }
 
+
+void testConvTransLayer(const string& type, bool trans, bool useGpu) {
+  TestConfig config;
+  config.biasSize = 3;
+  config.layerConfig.set_type(type);
+  config.layerConfig.set_num_filters(3);
+  config.layerConfig.set_partial_sum(1);
+  config.layerConfig.set_shared_biases(true);
+
+  config.inputDefs.push_back({INPUT_DATA, "layer_0", 1024, 288});
+  LayerInputConfig* input = config.layerConfig.add_inputs();
+  ConvConfig* conv = input->mutable_conv_conf();
+  conv->set_filter_size(2);
+  conv->set_filter_size_y(3);
+  conv->set_channels(16);
+  conv->set_padding(0);
+  conv->set_padding_y(1);
+  conv->set_stride(2);
+  conv->set_stride_y(2);
+  conv->set_groups(1);
+  conv->set_filter_channels(3 / conv->groups());
+  conv->set_img_size(16);
+  conv->set_output_x(
+      (2 * conv->padding() + conv->img_size() - conv->filter_size()) /
+          ((float)conv->stride()) +
+      1.5);
+
+  config.layerConfig.set_size(conv->img_size() * conv->img_size() *
+                              config.layerConfig.num_filters());
+
+  testLayerGrad(config, "convTrans", 100, trans, useGpu);
+}
+
+TEST(Layer, convTransLayer) {
+  testConvTransLayer("exconvt", /* trans= */ false, /* useGpu= */ false);
+/*
+#ifndef PADDLE_ONLY_CPU
+  testConvLayer("exconv",  trans=  false,  useGpu=  true);
+  testConvLayer("cudnn_conv",  trans=  false,  useGpu=  true);
+#endif
+*/
+}
+
 TEST(Layer, blockExpandLayer) {
   TestConfig config;
   config.biasSize = 0;