path

Author: zhanghang1989

.gitignore

@@ -6,5 +6,8 @@ build/
 data/
 docs/src/
 docs/html/
+encoding/lib/
 encoding/_ext/
 encoding.egg-info/
+experiments/recognition/
+experiments/segmentation/

LICENSE

@@ -9,8 +9,8 @@ to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
 copies of the Software, and to permit persons to whom the Software is
 furnished to do so, subject to the following conditions:
 
-The above copyright notice and this permission notice shall be included in all
-copies or substantial portions of the Software.
+The above copyright notice and this permission notice shall be included in
+all copies or substantial portions of the Software. 
 
 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,

build.py

@@ -16,6 +16,7 @@
 
 lib_path = os.path.join(os.path.dirname(torch.__file__), 'lib')
 cwd = os.path.dirname(os.path.realpath(__file__))
+encoding_lib_path = os.path.join(cwd, "encoding", "lib")
 
 # clean the build files
 clean_cmd = ['bash', 'clean.sh']
@@ -26,11 +27,12 @@
 if platform.system() == 'Darwin':
     os.environ['TH_LIBRARIES'] = os.path.join(lib_path,'libTH.1.dylib')
     os.environ['THC_LIBRARIES'] = os.path.join(lib_path,'libTHC.1.dylib')
-    ENCODING_LIB = os.path.join(lib_path, 'libENCODING.dylib')
+    ENCODING_LIB = os.path.join(cwd, 'encoding/lib/libENCODING.dylib')
+
 else:
     os.environ['TH_LIBRARIES'] = os.path.join(lib_path,'libTH.so.1')
     os.environ['THC_LIBRARIES'] = os.path.join(lib_path,'libTHC.so.1')
-    ENCODING_LIB = os.path.join(lib_path, 'libENCODING.so')
+    ENCODING_LIB = os.path.join(cwd, 'encoding/lib/libENCODING.so')
 
 build_all_cmd = ['bash', 'encoding/make.sh']
 subprocess.check_call(build_all_cmd, env=dict(os.environ))
@@ -44,7 +46,7 @@
 with_cuda = True 
 
 include_path = [os.path.join(lib_path, 'include'),
-                os.path.join(lib_path,'include/ENCODING'), 
+                os.path.join(cwd,'encoding/kernel'), 
                 os.path.join(cwd,'encoding/kernel/include'), 
                 os.path.join(cwd,'encoding/src/')]
 
@@ -65,6 +67,7 @@ def make_relative_rpath(path):
     include_dirs = include_path,
     extra_link_args = [
         make_relative_rpath(lib_path),
+        make_relative_rpath(encoding_lib_path),
         ENCODING_LIB,
     ],
 )

clean.sh

@@ -1,3 +1,3 @@
 #!/usr/bin/env bash
 
-rm -rf build/ dist/ encoding.egg-info/ encoding/build/ encoding/_ext/ __pycache__ encoding/__pycache__
+rm -rf build/ dist/ encoding.egg-info/ encoding/lib/ encoding/_ext/ __pycache__ encoding/__pycache__

docs/source/dilated.rst

@@ -4,11 +4,13 @@
 Dilated Networks
 ================
 
-We provide correct dilated pre-trained ResNet and DenseNet for semantic segmentation. 
-For dilation of ResNet, we replace the stride of 2 Conv3x3 at begining of certain stage and update the dilation of the conv layers afterwards. 
-For dilation of DenseNet, we provide :class:`encoding.nn.DilatedAvgPool2d` that handles the dilation of the transition layers, then update the dilation of the conv layers afterwards. 
+We provide correct dilated pre-trained ResNet and DenseNet (stride of 8) for semantic segmentation. 
+For dilation of DenseNet, we provide :class:`encoding.nn.DilatedAvgPool2d`. 
 All provided models have been verified. 
 
+.. note::
+
+    This code is provided together with the paper (coming soon), please cite our work.
 
 .. automodule:: encoding.dilated
 .. currentmodule:: encoding.dilated

docs/source/notes/compile.rst

@@ -5,9 +5,7 @@ Installing PyTorch-Encoding
 Install from Source
 -------------------
 
-    * Please follow the `PyTorch instructions <https://github.com/pytorch/pytorch#from-source>`_ to install PyTorch from Source to the ``$HOME`` directory (recommended). Or you can simply clone a copy to ``$HOME`` directory::
-
-        git clone https://github.com/pytorch/pytorch $HOME/pytorch
+    * Install PyTorch from Source (recommended). Please follow the `PyTorch instructions <https://github.com/pytorch/pytorch#from-source>`_.
 
     * Install this package
 

docs/source/utils.rst

@@ -9,10 +9,10 @@ Useful util functions.
 .. automodule:: encoding.utils
 .. currentmodule:: encoding.utils
 
-:hidden:`CosLR_Scheduler`
+:hidden:`LR_Scheduler`
 ~~~~~~~~~~~~~~~~~~~~~~~~~
 
-.. autoclass:: CosLR_Scheduler
+.. autoclass:: LR_Scheduler
     :members:
 
 :hidden:`get_optimizer`

encoding/CMakeLists.txt

@@ -75,9 +75,4 @@ IF(ENCODING_SO_VERSION)
     SOVERSION ${ENCODING_SO_VERSION})
 ENDIF(ENCODING_SO_VERSION)
 
-FILE(GLOB kernel-header kernel/generic/*.h)
-FILE(GLOB src-header src/generic/*.h)
-
-INSTALL(TARGETS ENCODING LIBRARY DESTINATION ${ENCODING_INSTALL_LIB_SUBDIR})
-INSTALL(FILES kernel/thc_encoding.h DESTINATION "${ENCODING_INSTALL_INCLUDE_SUBDIR}/ENCODING")
-INSTALL(FILES ${src-header} ${kernel-header} DESTINATION "${ENCODING_INSTALL_INCLUDE_SUBDIR}/ENCODING/generic")
+#INSTALL(TARGETS ENCODING LIBRARY DESTINATION ${ENCODING_INSTALL_LIB_SUBDIR})

encoding/functions/syncbn.py

@@ -61,7 +61,11 @@ def sum_square(input):
     return _sum_square()(input)
 
 
-class _batchnormtrain(Function):
+class _batchnorm(Function):
+    def __init__(self, training=False):
+        super(_batchnorm, self).__init__()
+        self.training = training
+
     def forward(ctx, input, gamma, beta, mean, std):
         ctx.save_for_backward(input, gamma, beta, mean, std)
         assert(input.dim()==3)
@@ -95,13 +99,13 @@ def backward(ctx, gradOutput):
                 encoding_lib.Encoding_Float_batchnorm_Backward(
                     gradOutput, input, gradInput, gradGamma, gradBeta, 
                     mean, invstd, gamma, beta, gradMean, gradStd,
-                    True) 
+                    self.training) 
         elif isinstance(input, torch.cuda.DoubleTensor):
             with torch.cuda.device_of(input):
                 encoding_lib.Encoding_Double_batchnorm_Backward(
                     gradOutput, input, gradInput, gradGamma, gradBeta, 
                     mean, invstd, gamma, beta, gradMean, gradStd,
-                    True) 
+                    self.training) 
         else:
             raise RuntimeError('Unimplemented data type!')
         return gradInput, gradGamma, gradBeta, gradMean, gradStd
@@ -122,52 +126,7 @@ def batchnormtrain(input, gamma, beta, mean, std):
         - Output: :math:`(N, C)` or :math:`(N, C, L)` (same shape as input)
 
     """
-    return _batchnormtrain()(input, gamma, beta, mean, std)
-
-
-class _batchnormeval(Function):
-    def forward(ctx, input, gamma, beta, mean, std):
-        ctx.save_for_backward(input, gamma, beta, mean, std)
-        assert(input.dim()==3)
-        with torch.cuda.device_of(input):
-            invstd = 1.0 / std
-            output = input.new().resize_as_(input)
-        if isinstance(input, torch.cuda.FloatTensor):
-            with torch.cuda.device_of(input):
-                encoding_lib.Encoding_Float_batchnorm_Forward(output, 
-                    input, mean, invstd, gamma, beta)
-        elif isinstance(input, torch.cuda.DoubleTensor):
-            with torch.cuda.device_of(input):
-                encoding_lib.Encoding_Double_batchnorm_Forward(output, 
-                    input, mean, invstd, gamma, beta)
-        else:
-            raise RuntimeError('Unimplemented data type!')
-        return output 
-
-    def backward(ctx, gradOutput):
-        input, gamma, beta, mean, std = ctx.saved_tensors
-        invstd = 1.0 / std
-        with torch.cuda.device_of(input):
-            gradInput = gradOutput.new().resize_as_(input).zero_()
-            gradGamma = gradOutput.new().resize_as_(gamma).zero_()
-            gradBeta  = gradOutput.new().resize_as_(beta).zero_()
-            gradMean  = gradOutput.new().resize_as_(mean).zero_()
-            gradStd   = gradOutput.new().resize_as_(std).zero_()
-        if isinstance(input, torch.cuda.FloatTensor):
-            with torch.cuda.device_of(input):
-                encoding_lib.Encoding_Float_batchnorm_Backward(
-                    gradOutput, input, gradInput, gradGamma, gradBeta, 
-                    mean, invstd, gamma, beta, gradMean, gradStd,
-                    False) 
-        elif isinstance(input, torch.cuda.DoubleTensor):
-            with torch.cuda.device_of(input):
-                encoding_lib.Encoding_Double_batchnorm_Backward(
-                    gradOutput, input, gradInput, gradGamma, gradBeta, 
-                    mean, invstd, gamma, beta, gradMean, gradStd,
-                    False) 
-        else:
-            raise RuntimeError('Unimplemented data type!')
-        return gradInput, gradGamma, gradBeta, gradMean, gradStd
+    return _batchnorm(True)(input, gamma, beta, mean, std)
 
 
 def batchnormeval(input, gamma, beta, mean, std):
@@ -176,4 +135,4 @@ def batchnormeval(input, gamma, beta, mean, std):
 
     Please see encoding.batchnormtrain_
     """
-    return _batchnormeval()(input, gamma, beta, mean, std)
+    return _batchnorm(False)(input, gamma, beta, mean, std)

encoding/make.sh

@@ -1,7 +1,6 @@
 #!/usr/bin/env bash
 
-mkdir -p encoding/build && cd encoding/build
+mkdir -p encoding/lib && cd encoding/lib
 # compile and install
 cmake ..
-make install
-cd ..
+make