[OP] 3d conv + 3d pool (#2585)

* cudnn 3d conv & 3d pooling
starimpact · Jul 8, 2016 · 5532511 · 5532511
1 parent 70fe575
commit 5532511
Show file tree

Hide file tree

Showing 27 changed files with 692 additions and 314 deletions.
diff --git a/Makefile b/Makefile
@@ -36,9 +36,9 @@ endif
 CFLAGS += -I$(ROOTDIR)/mshadow/ -I$(ROOTDIR)/dmlc-core/include -fPIC -Iinclude $(MSHADOW_CFLAGS)
 LDFLAGS = -pthread $(MSHADOW_LDFLAGS) $(DMLC_LDFLAGS)
 ifeq ($(DEBUG), 1)
-	NVCCFLAGS = -Xcompiler -std=c++98 -D_FORCE_INLINES -g -G -O0 -ccbin $(CXX) $(MSHADOW_NVCCFLAGS)
+	NVCCFLAGS = -std=c++11 -Xcompiler -D_FORCE_INLINES -g -G -O0 -ccbin $(CXX) $(MSHADOW_NVCCFLAGS)
 else
-	NVCCFLAGS = -Xcompiler -std=c++98 -D_FORCE_INLINES -g -O3 -ccbin $(CXX) $(MSHADOW_NVCCFLAGS)
+	NVCCFLAGS = -std=c++11 -Xcompiler -D_FORCE_INLINES -g -O3 -ccbin $(CXX) $(MSHADOW_NVCCFLAGS)
 endif
 
 ifndef LINT_LANG
@@ -149,8 +149,8 @@ endif
 
 build/src/%.o: src/%.cc
 	@mkdir -p $(@D)
-	$(CXX) -std=c++0x $(CFLAGS) -MM -MT build/src/$*.o $< >build/src/$*.d
-	$(CXX) -std=c++0x -c $(CFLAGS) -c $< -o $@
+	$(CXX) -std=c++11 $(CFLAGS) -MM -MT build/src/$*.o $< >build/src/$*.d
+	$(CXX) -std=c++11 -c $(CFLAGS) -c $< -o $@
 
 build/src/%_gpu.o: src/%.cu
 	@mkdir -p $(@D)
@@ -159,20 +159,20 @@ build/src/%_gpu.o: src/%.cu
 
 build/plugin/%.o: plugin/%.cc
 	@mkdir -p $(@D)
-	$(CXX) -std=c++0x $(CFLAGS) -MM -MT build/plugin/$*.o $< >build/plugin/$*.d
-	$(CXX) -std=c++0x -c $(CFLAGS) -c $< -o $@
+	$(CXX) -std=c++11 $(CFLAGS) -MM -MT build/plugin/$*.o $< >build/plugin/$*.d
+	$(CXX) -std=c++11 -c $(CFLAGS) -c $< -o $@
 
 # A nvcc bug cause it to generate "generic/xxx.h" dependencies from torch headers.
 # Use CXX to generate dependency instead.
 build/plugin/%_gpu.o: plugin/%.cu
 	@mkdir -p $(@D)
-	$(CXX) -std=c++0x $(CFLAGS) -MM -MT build/plugin/$*_gpu.o $< >build/plugin/$*_gpu.d
+	$(CXX) -std=c++11 $(CFLAGS) -MM -MT build/plugin/$*_gpu.o $< >build/plugin/$*_gpu.d
 	$(NVCC) -c -o $@ $(NVCCFLAGS) -Xcompiler "$(CFLAGS)" $<
 
 $(EXTRA_OPERATORS)/build/%.o: $(EXTRA_OPERATORS)/%.cc
 	@mkdir -p $(@D)
-	$(CXX) -std=c++0x $(CFLAGS) -Isrc/operator -MM -MT $(EXTRA_OPERATORS)/build/$*.o $< >$(EXTRA_OPERATORS)/build/$*.d
-	$(CXX) -std=c++0x -c $(CFLAGS) -Isrc/operator -c $< -o $@
+	$(CXX) -std=c++11 $(CFLAGS) -Isrc/operator -MM -MT $(EXTRA_OPERATORS)/build/$*.o $< >$(EXTRA_OPERATORS)/build/$*.d
+	$(CXX) -std=c++11 -c $(CFLAGS) -Isrc/operator -c $< -o $@
 
 $(EXTRA_OPERATORS)/build/%_gpu.o: $(EXTRA_OPERATORS)/%.cu
 	@mkdir -p $(@D)
@@ -200,7 +200,7 @@ bin/im2rec: tools/im2rec.cc $(ALL_DEP)
 
 $(BIN) :
 	@mkdir -p $(@D)
-	$(CXX) $(CFLAGS) -std=c++0x  -o $@ $(filter %.cpp %.o %.c %.a %.cc, $^) $(LDFLAGS)
+	$(CXX) $(CFLAGS) -std=c++11  -o $@ $(filter %.cpp %.o %.c %.a %.cc, $^) $(LDFLAGS)
 
 include tests/cpp/unittest.mk
 

diff --git a/R-package/vignettes/classifyRealImageWithPretrainedModel.Rmd b/R-package/vignettes/classifyRealImageWithPretrainedModel.Rmd
@@ -7,7 +7,7 @@ algorithm can do is to classify real world images.
 In this example we will show how to use a pretrained Inception-BatchNorm Network to predict the class of
 real world image. The network architecture is decribed in [1].
 
-The pre-trained Inception-BatchNorm network is able to be downloaded from [this link](http://webdocs.cs.ualberta.ca/~bx3/data/Inception.zip)
+The pre-trained Inception-BatchNorm network is able to be downloaded from [this link](http://data.dmlc.ml/mxnet/data/Inception.zip)
 This model gives the recent state-of-art prediction accuracy on image net dataset.
 
 Preface
@@ -69,7 +69,7 @@ preproc.image <- function(im, mean.image) {
   shape <- dim(im)
   short.edge <- min(shape[1:2])
   xx <- floor((shape[1] - short.edge) / 2)
-  yy <- floor((shape[2] - short.edge) / 2) 
+  yy <- floor((shape[2] - short.edge) / 2)
   croped <- crop.borders(im, xx, yy)
   # resize to 224 x 224, needed by input of the model.
   resized <- resize(croped, 224, 224)

diff --git a/dmlc-core b/dmlc-core
diff --git a/docs/how_to/cloud.md b/docs/how_to/cloud.md
@@ -29,7 +29,7 @@ There are several ways to upload local data to S3. One simple way is using
 [s3cmd](http://s3tools.org/s3cmd). For example:
 
 ```bash
-wget http://webdocs.cs.ualberta.ca/~bx3/data/mnist.zip
+wget http://data.dmlc.ml/mxnet/data/mnist.zip
 unzip mnist.zip && s3cmd put t*-ubyte s3://dmlc/mnist/
 ```
 

diff --git a/docs/packages/r/classifyRealImageWithPretrainedModel.md b/docs/packages/r/classifyRealImageWithPretrainedModel.md
@@ -6,7 +6,7 @@ algorithm can do is to classify real world images.
 In this example we will show how to use a pretrained Inception-BatchNorm Network to predict the class of
 real world image. The network architecture is decribed in [1].
 
-The pre-trained Inception-BatchNorm network is able to be downloaded from [this link](http://webdocs.cs.ualberta.ca/~bx3/data/Inception.zip)
+The pre-trained Inception-BatchNorm network is able to be downloaded from [this link](http://data.dmlc.ml/mxnet/data/Inception.zip)
 This model gives the recent state-of-art prediction accuracy on image net dataset.
 
 Preface
@@ -112,7 +112,7 @@ preproc.image <- function(im, mean.image) {
   shape <- dim(im)
   short.edge <- min(shape[1:2])
   xx <- floor((shape[1] - short.edge) / 2)
-  yy <- floor((shape[2] - short.edge) / 2) 
+  yy <- floor((shape[2] - short.edge) / 2)
   croped <- crop.borders(im, xx, yy)
   # resize to 224 x 224, needed by input of the model.
   resized <- resize(croped, 224, 224)

diff --git a/example/image-classification/symbol_inception-bn-28-small.py b/example/image-classification/symbol_inception-bn-28-small.py
@@ -17,7 +17,7 @@ def DownsampleFactory(data, ch_3x3, mirror_attr):
     # conv 3x3
     conv = ConvFactory(data=data, kernel=(3, 3), stride=(2, 2), num_filter=ch_3x3, pad=(1, 1), mirror_attr=mirror_attr)
     # pool
-    pool = mx.symbol.Pooling(data=data, kernel=(3, 3), stride=(2, 2), pool_type='max', attr=mirror_attr)
+    pool = mx.symbol.Pooling(data=data, kernel=(3, 3), stride=(2, 2), pad=(1, 1), pool_type='max', attr=mirror_attr)
     # concat
     concat = mx.symbol.Concat(*[conv, pool])
     return concat

diff --git a/example/image-classification/symbol_inception-bn-full.py b/example/image-classification/symbol_inception-bn-full.py
@@ -37,7 +37,7 @@ def InceptionFactoryB(data, num_3x3red, num_3x3, num_d3x3red, num_d3x3, name):
     cd3x3 = ConvFactory(data=cd3x3r, num_filter=num_d3x3, kernel=(3, 3), pad=(1, 1), stride=(1, 1), name=('%s_double_3x3_0' % name))
     cd3x3 = ConvFactory(data=cd3x3, num_filter=num_d3x3, kernel=(3, 3), pad=(1, 1), stride=(2, 2), name=('%s_double_3x3_1' % name))
     # pool + proj
-    pooling = mx.symbol.Pooling(data=data, kernel=(3, 3), stride=(2, 2), pool_type="max", name=('max_pool_%s_pool' % name))
+    pooling = mx.symbol.Pooling(data=data, kernel=(3, 3), stride=(2, 2), pad=(1, 1), pool_type="max", name=('max_pool_%s_pool' % name))
     # concat
     concat = mx.symbol.Concat(*[c3x3, cd3x3, pooling], name='ch_concat_%s_chconcat' % name)
     return concat

diff --git a/example/image-classification/symbol_inception-bn.py b/example/image-classification/symbol_inception-bn.py
@@ -45,7 +45,7 @@ def InceptionFactoryB(data, num_3x3red, num_3x3, num_d3x3red, num_d3x3, name):
     cd3x3 = ConvFactory(data=cd3x3r, num_filter=num_d3x3, kernel=(3, 3), pad=(1, 1), stride=(1, 1), name=('%s_double_3x3_0' % name))
     cd3x3 = ConvFactory(data=cd3x3, num_filter=num_d3x3, kernel=(3, 3), pad=(1, 1), stride=(2, 2), name=('%s_double_3x3_1' % name))
     # pool + proj
-    pooling = mx.symbol.Pooling(data=data, kernel=(3, 3), stride=(2, 2), pool_type="max", name=('max_pool_%s_pool' % name))
+    pooling = mx.symbol.Pooling(data=data, kernel=(3, 3), stride=(2, 2), pad=(1, 1), pool_type="max", name=('max_pool_%s_pool' % name))
     # concat
     concat = mx.symbol.Concat(*[c3x3, cd3x3, pooling], name='ch_concat_%s_chconcat' % name)
     return concat

diff --git a/example/image-classification/train_cifar10.py b/example/image-classification/train_cifar10.py
@@ -40,7 +40,7 @@ def _download(data_dir):
     os.chdir(data_dir)
     if (not os.path.exists('train.rec')) or \
        (not os.path.exists('test.rec')) :
-        os.system("wget http://webdocs.cs.ualberta.ca/~bx3/data/cifar10.zip")
+        os.system("wget http://data.dmlc.ml/mxnet/data/cifar10.zip")
         os.system("unzip -u cifar10.zip")
         os.system("mv cifar/* .; rm -rf cifar; rm cifar10.zip")
     os.chdir("..")

diff --git a/example/image-classification/train_cifar10_mirroring.py b/example/image-classification/train_cifar10_mirroring.py
@@ -8,7 +8,7 @@
 # documentation could be expected when this feature is mature.
 #
 # When mirroring is turned on and set properly, we could expect smaller memory
-# consumption with slightly slower computation speed (due to extra forward 
+# consumption with slightly slower computation speed (due to extra forward
 # steps). We are not including a sample running log here, as this test case
 # is only a functionality test. The using of pycuda GPU memory query is also
 # not very good way of measuring the memory usage here.
@@ -55,7 +55,7 @@ def _download(data_dir):
     os.chdir(data_dir)
     if (not os.path.exists('train.rec')) or \
        (not os.path.exists('test.rec')) :
-        os.system("wget http://webdocs.cs.ualberta.ca/~bx3/data/cifar10.zip")
+        os.system("wget http://data.dmlc.ml/mxnet/data/data/cifar10.zip")
         os.system("unzip -u cifar10.zip")
         os.system("mv cifar/* .; rm -rf cifar; rm cifar10.zip")
     os.chdir("..")

diff --git a/example/image-classification/train_cifar10_resnet.py b/example/image-classification/train_cifar10_resnet.py
@@ -79,7 +79,7 @@ def _download(data_dir):
     os.chdir(data_dir)
     if (not os.path.exists('train.rec')) or \
        (not os.path.exists('test.rec')):
-        os.system('wget http://webdocs.cs.ualberta.ca/~bx3/data/cifar10.zip')
+        os.system('wget http://data.dmlc.ml/mxnet/data/cifar10.zip')
         os.system('unzip -u cifar10.zip')
         os.system('mv cifar/* .; rm -rf cifar; rm cifar10.zip')
     os.chdir('..')

diff --git a/example/image-classification/train_mnist.R b/example/image-classification/train_mnist.R
@@ -4,11 +4,11 @@ require(mxnet)
 download_ <- function(data_dir) {
     dir.create(data_dir, showWarnings = FALSE)
     setwd(data_dir)
-    if ((!file.exists('train-images-idx3-ubyte')) || 
+    if ((!file.exists('train-images-idx3-ubyte')) ||
         (!file.exists('train-labels-idx1-ubyte')) ||
         (!file.exists('t10k-images-idx3-ubyte')) ||
         (!file.exists('t10k-labels-idx1-ubyte'))) {
-        download.file(url='http://webdocs.cs.ualberta.ca/~bx3/data/mnist.zip',
+        download.file(url='http://data.dmlc.ml/mxnet/data/mnist.zip',
                       destfile='mnist.zip', method='wget')
         unzip("mnist.zip")
         file.remove("mnist.zip")
@@ -83,7 +83,7 @@ get_iterator <- function(data_shape) {
 }
 
 parse_args <- function() {
-    parser <- ArgumentParser(description='train an image classifer on mnist')  
+    parser <- ArgumentParser(description='train an image classifer on mnist')
     parser$add_argument('--network', type='character', default='mlp',
                         choices = c('mlp', 'lenet'),
                         help = 'the cnn to use')

diff --git a/example/image-classification/train_mnist.py b/example/image-classification/train_mnist.py
@@ -12,7 +12,7 @@ def _download(data_dir):
        (not os.path.exists('train-labels-idx1-ubyte')) or \
        (not os.path.exists('t10k-images-idx3-ubyte')) or \
        (not os.path.exists('t10k-labels-idx1-ubyte')):
-        os.system("wget http://webdocs.cs.ualberta.ca/~bx3/data/mnist.zip")
+        os.system("wget http://data.dmlc.ml/mxnet/data/mnist.zip")
         os.system("unzip -u mnist.zip; rm mnist.zip")
     os.chdir("..")
 

diff --git a/example/module/train_cifar10.py b/example/module/train_cifar10.py
@@ -52,7 +52,7 @@ def _download(data_dir):
     os.chdir(data_dir)
     if (not os.path.exists('train.rec')) or \
        (not os.path.exists('test.rec')) :
-        os.system("wget http://webdocs.cs.ualberta.ca/~bx3/data/cifar10.zip")
+           os.system("wget http://data.dmlc.ml/mxnet/data/cifar10.zip")
         os.system("unzip -u cifar10.zip")
         os.system("mv cifar/* .; rm -rf cifar; rm cifar10.zip")
     os.chdir(cwd)

diff --git a/example/notebooks/predict-with-pretrained-model.ipynb b/example/notebooks/predict-with-pretrained-model.ipynb
@@ -16,7 +16,7 @@
     "For network structure, you can visualize it in [Composite Symbol Demo](composite_symbol.ipynb)\n",
     "\n",
     "The pre-trained Inception-BatchNorm network is able to be downloaded from:\n",
-    "[http://webdocs.cs.ualberta.ca/~bx3/data/Inception.zip](http://webdocs.cs.ualberta.ca/~bx3/data/Inception.zip)\n",
+    "[http://data.dmlc.ml/mxnet/data/Inception.zip](http://data.dmlc.ml/mxnet/data/Inception.zip)\n",
     "This model achieves Top-1 Accuracy: 70% and Top-5 Accuracy: 89.9%\n",
     "\n",
     "Note: This network is trained by using very simple augmentation (random flip + random crop). We will release model with a little bit more augmentation (which achieves better validation score)"

diff --git a/example/rnn/char-rnn.ipynb b/example/rnn/char-rnn.ipynb
@@ -43,14 +43,14 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "<img src=\"http://webdocs.cs.ualberta.ca/~bx3/char-rnn_1.png\">\n"
+    "<img src=\"http://data.dmlc.ml/mxnet/data/char-rnn_1.png\">\n"
    ]
   },
   {
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "<img src=\"http://webdocs.cs.ualberta.ca/~bx3/char-rnn_2.png\">"
+    "<img src=\"http://data.dmlc.ml/mxnet/data/char-rnn_2.png\">"
    ]
   },
   {
@@ -145,7 +145,7 @@
    ],
    "source": [
     "import os\n",
-    "data_url = \"http://webdocs.cs.ualberta.ca/~bx3/lab_data.zip\"\n",
+    "data_url = \"http://data.dmlc.ml/mxnet/data/lab_data.zip\"\n",
     "os.system(\"wget %s\" % data_url)\n",
     "os.system(\"unzip -o lab_data.zip\")"
    ]

diff --git a/scala-package/core/scripts/get_cifar_data.sh b/scala-package/core/scripts/get_cifar_data.sh
@@ -5,7 +5,7 @@ fi
 
 cifar_data_path="./data/cifar10.zip"
 if [ ! -f "$cifar_data_path" ]; then
-  wget http://webdocs.cs.ualberta.ca/~bx3/data/cifar10.zip -P $data_path
+  wget http://data.dmlc.ml/mxnet/data/cifar10.zip -P $data_path
   cd $data_path
   unzip -u cifar10.zip
-fi
+fi
diff --git a/scala-package/core/scripts/get_mnist_data.sh b/scala-package/core/scripts/get_mnist_data.sh
@@ -5,7 +5,7 @@ fi
 
 mnist_data_path="./data/mnist.zip"
 if [ ! -f "$mnist_data_path" ]; then
-  wget http://webdocs.cs.ualberta.ca/~bx3/data/mnist.zip -P $data_path
+  wget http://data.dmlc.ml/mxnet/data/mnist.zip -P $data_path
   cd $data_path
   unzip -u mnist.zip
 fi
diff --git a/scala-package/core/src/test/scala/ml/dmlc/mxnet/train/ConvSuite.scala b/scala-package/core/src/test/scala/ml/dmlc/mxnet/train/ConvSuite.scala
@@ -91,6 +91,6 @@ class ConvSuite extends FunSuite with BeforeAndAfterAll {
     }
     val acc = numCorrect.toFloat / numInst
     logger.info(s"Final accuracy = $acc")
-    assert(acc > 0.96)
+    assert(acc > 0.95)
   }
 }