Skip to content

GitLab

".github/vscode:/vscode.git/clone" did not exist on "1ed0f6f614fb1cf44a3d86cfa6ed0701a0c5a14d"
Commit 2d21747a authored by Zhang's avatar Zhang
Browse files

v0.4.2

parent 7e19143c
Hide whitespace changes
Inline Side-by-side
Showing with 897 additions and 547 deletions
build.py deleted 100644 → 0
View file @ 7e19143c
##+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
## Created by: Hang Zhang
## ECE Department, Rutgers University
## Email: zhang.hang@rutgers.edu
## Copyright (c) 2017
##
## This source code is licensed under the MIT-style license found in the
## LICENSE file in the root directory of this source tree
##+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
import os
import torch
import platform
import subprocess
from torch.utils.ffi import create_extension
torch_ver = torch.__version__[:3]
lib_path = os.path.join(os.path.dirname(torch.__file__), 'lib')
cwd = os.path.join(os.path.dirname(os.path.realpath(__file__)), 'encoding/')
encoding_lib_path = os.path.join(cwd, "lib")
# clean the build files
clean_cmd = ['bash', 'clean.sh']
subprocess.check_call(clean_cmd)
# build CUDA library
os.environ['TORCH_BUILD_DIR'] = lib_path
if platform.system() == 'Darwin':
if torch_ver == '0.3':
os.environ['TH_LIBRARIES'] = os.path.join(lib_path,'libATen.1.dylib')
else:
os.environ['TH_LIBRARIES'] = os.path.join(lib_path,'libATen.dylib')
ENCODING_LIB = os.path.join(cwd, 'lib/libENCODING.dylib')
else:
os.environ['CFLAGS'] = '-std=c99'
if torch_ver == '0.3':
os.environ['TH_LIBRARIES'] = os.path.join(lib_path,'libATen.so.1')
else:
os.environ['TH_LIBRARIES'] = os.path.join(lib_path,'libATen.so')
ENCODING_LIB = os.path.join(cwd, 'lib/libENCODING.so')
build_all_cmd = ['bash', 'encoding/make.sh']
subprocess.check_call(build_all_cmd, env=dict(os.environ))
# build FFI
sources = ['encoding/src/encoding_lib.cpp']
headers = [
'encoding/src/encoding_lib.h',
]
defines = [('WITH_CUDA', None)]
with_cuda = True
include_path = [os.path.join(lib_path, 'include'),
os.path.join(cwd,'kernel'),
os.path.join(cwd,'kernel/include'),
os.path.join(cwd,'src/')]
def make_relative_rpath(path):
if platform.system() == 'Darwin':
return '-Wl,-rpath,' + path
else:
return '-Wl,-rpath,' + path
ffi = create_extension(
'encoding._ext.encoding_lib',
package=True,
headers=headers,
sources=sources,
define_macros=defines,
relative_to=__file__,
with_cuda=with_cuda,
extra_compile_args=["-std=c99"],
include_dirs = include_path,
extra_link_args = [
make_relative_rpath(lib_path),
make_relative_rpath(encoding_lib_path),
ENCODING_LIB,
],
)
if __name__ == '__main__':
ffi.build()
docs/source/_templates/layout.html 0 → 100644
View file @ 2d21747a
{% extends "!layout.html" %}
{%- block extrahead %}
<script type="text/javascript" src="http://zhanghang1989.github.io/files/hidebib.js"></script>
{% endblock %}
docs/source/dilated.rst
View file @ 2d21747a
...@@ -50,37 +50,3 @@ ResNet ...@@ -50,37 +50,3 @@ ResNet
~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~
.. autofunction:: resnet152 .. autofunction:: resnet152
DenseNet
--------
:hidden:`DenseNet`
~~~~~~~~~~~~~~~~~~
.. autoclass:: DenseNet
:members:
:hidden:`densenet161`
~~~~~~~~~~~~~~~~~~~~~
.. autofunction:: densenet161
:hidden:`densenet121`
~~~~~~~~~~~~~~~~~~~~~
.. autofunction:: densenet121
:hidden:`densenet169`
~~~~~~~~~~~~~~~~~~~~~
.. autofunction:: densenet169
:hidden:`densenet201`
~~~~~~~~~~~~~~~~~~~~~
.. autofunction:: densenet201
docs/source/encoding.rst deleted 100644 → 0
View file @ 7e19143c
.. role:: hidden
:class: hidden-section
NN Layers
=========
Customized NN modules in Encoding Package. For Synchronized Cross-GPU Batch Normalization, please visit :class:`encoding.nn.BatchNorm2d`.
.. currentmodule:: encoding.nn
:hidden:`Encoding`
~~~~~~~~~~~~~~~~~~
.. autoclass:: Encoding
:members:
:hidden:`Inspiration`
~~~~~~~~~~~~~~~~~~~~~
.. autoclass:: Inspiration
:members:
:hidden:`UpsampleConv2d`
~~~~~~~~~~~~~~~~~~~~~~~~~~
.. autoclass:: UpsampleConv2d
:members:
:hidden:`DilatedAvgPool2d`
~~~~~~~~~~~~~~~~~~~~~~~~~~
.. autoclass:: DilatedAvgPool2d
:members:
:hidden:`GramMatrix`
~~~~~~~~~~~~~~~~~~~~
.. autoclass:: GramMatrix
:members:
docs/source/experiments/segmentation.rst 0 → 100644
View file @ 2d21747a
Context Encoding for Semantic Segmentation (EncNet)
===================================================
Install Package
---------------
- Clone the GitHub repo::
git clone git@github.com:zhanghang1989/PyTorch-Encoding.git
- Install PyTorch Encoding (if not yet). Please follow the installation guide `Installing PyTorch Encoding <../notes/compile.html>`_.
Test Pre-trained Model
----------------------
.. hint::
The model names contain the training information. For instance ``FCN_ResNet50_PContext``:
- ``FCN`` indicate the algorithm is “Fully Convolutional Network for Semantic Segmentation”
- ``ResNet50`` is the name of backbone network.
- ``PContext`` means the PASCAL in Context dataset.
How to get pretrained model, for example ``FCN_ResNet50_PContext``::
model = encoding.models.get_model('FCN_ResNet50_PContext', pretrained=True)
The test script is in the ``experiments/segmentation/`` folder. For evaluating the model (using MS),
for example ``Encnet_ResNet50_PContext``::
python test.py --dataset PContext --model-zoo Encnet_ResNet50_PContext --eval
# pixAcc: 0.7862, mIoU: 0.4946: 100%|████████████████████████| 319/319 [09:44<00:00, 1.83s/it]
The command for training the model can be found by clicking ``cmd`` in the table.
.. role:: raw-html(raw)
:format: html
+----------------------------------+-----------+-----------+---------------------------------------------------------------------------------------------+
| Model | pixAcc | mIoU | Command |
+==================================+===========+===========+=============================================================================================+
| FCN_ResNet50_PContext | 76.0% | 45.7 | :raw-html:`<a href="javascript:toggleblock('cmd_fcn50_pcont')" class="toggleblock">cmd</a>` |
+----------------------------------+-----------+-----------+---------------------------------------------------------------------------------------------+
| Encnet_ResNet50_PContext | 78.6% | 49.5 | :raw-html:`<a href="javascript:toggleblock('cmd_enc50_pcont')" class="toggleblock">cmd</a>` |
+----------------------------------+-----------+-----------+---------------------------------------------------------------------------------------------+
.. raw:: html
<code xml:space="preserve" id="cmd_fcn50_pcont" style="display: none; text-align: left; white-space: pre-wrap">
CUDA_VISIBLE_DEVICES=0,1,2,3 python train.py --dataset PContext --model FCN
</code>
<code xml:space="preserve" id="cmd_enc50_pcont" style="display: none; text-align: left; white-space: pre-wrap">
CUDA_VISIBLE_DEVICES=0,1,2,3 python train.py --dataset PContext --model EncNet --aux --se-loss
</code>
Quick Demo
~~~~~~~~~~
.. code-block:: python
import torch
import encoding
# Get the model
model = encoding.models.get_model('Encnet_ResNet50_PContext', pretrained=True).cuda()
model.eval()
# Prepare the image
url = 'https://github.com/zhanghang1989/image-data/blob/master/' + \
'encoding/segmentation/pcontext/2010_001829_org.jpg?raw=true'
filename = 'example.jpg'
img = encoding.utils.load_image(
encoding.utils.download(url, filename)).cuda().unsqueeze(0)
# Make prediction
output = model.evaluate(img)
predict = torch.max(output, 1)[1].cpu().numpy() + 1
# Get color pallete for visualization
mask = encoding.utils.get_mask_pallete(predict, 'pcontext')
mask.save('output.png')
.. image:: https://raw.githubusercontent.com/zhanghang1989/image-data/master/encoding/segmentation/pcontext/2010_001829_org.jpg
:width: 45%
.. image:: https://raw.githubusercontent.com/zhanghang1989/image-data/master/encoding/segmentation/pcontext/2010_001829.png
:width: 45%
Train Your Own Model
--------------------
- Prepare the datasets by runing the scripts in the ``scripts/`` folder, for example preparing ``PASCAL Context`` dataset::
python scripts/prepare_pcontext.py
- The training script is in the ``experiments/segmentation/`` folder, example training command::
CUDA_VISIBLE_DEVICES=0,1,2,3 python train.py --dataset pcontext --model encnet --aux --se-loss
- Detail training options, please run ``python train.py -h``.
Citation
--------
.. note::
* Hang Zhang, Kristin Dana, Jianping Shi, Zhongyue Zhang, Xiaogang Wang, Ambrish Tyagi, Amit Agrawal. "Context Encoding for Semantic Segmentation" *The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2018*::
@InProceedings{Zhang_2018_CVPR,
author = {Zhang, Hang and Dana, Kristin and Shi, Jianping and Zhang, Zhongyue and Wang, Xiaogang and Tyagi, Ambrish and Agrawal, Amit},
title = {Context Encoding for Semantic Segmentation},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2018}
}
docs/source/functions.rst
View file @ 2d21747a
...@@ -9,19 +9,10 @@ encoding.functions ...@@ -9,19 +9,10 @@ encoding.functions
.. currentmodule:: encoding.functions .. currentmodule:: encoding.functions
:hidden:`batchnorm` :hidden:`batchnormtrain`
~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~
.. autofunction:: batchnorm .. autofunction:: batchnormtrain
:hidden:`batchnormeval`
~~~~~~~~~~~~~~~~~~~~~~~
.. autofunction:: batchnormeval
:hidden:`dilatedavgpool2d`
~~~~~~~~~~~~~~~~~~~~~~~~~~
.. autofunction:: dilatedavgpool2d
:hidden:`aggregate` :hidden:`aggregate`
~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~
......
docs/source/nn.rst
View file @ 2d21747a
...@@ -4,7 +4,7 @@ ...@@ -4,7 +4,7 @@
encoding.nn encoding.nn
=========== ===========
Customized NN modules in Encoding Package. For Synchronized Cross-GPU Batch Normalization, please visit :class:`encoding.nn.SyncBatchNorm2d`. Customized NN modules in Encoding Package. For Synchronized Cross-GPU Batch Normalization, please visit :class:`encoding.nn.BatchNorm2d`.
.. currentmodule:: encoding.nn .. currentmodule:: encoding.nn
...@@ -14,22 +14,22 @@ Customized NN modules in Encoding Package. For Synchronized Cross-GPU Batch Norm ...@@ -14,22 +14,22 @@ Customized NN modules in Encoding Package. For Synchronized Cross-GPU Batch Norm
.. autoclass:: Encoding .. autoclass:: Encoding
:members: :members:
:hidden:`SyncBatchNorm2d` :hidden:`BatchNorm2d`
~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~
.. autoclass:: SyncBatchNorm2d .. autoclass:: BatchNorm2d
:members: :members:
:hidden:`SyncBatchNorm1d` :hidden:`BatchNorm1d`
~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~
.. autoclass:: SyncBatchNorm1d .. autoclass:: BatchNorm1d
:members: :members:
:hidden:`SyncBatchNorm3d` :hidden:`BatchNorm3d`
~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~
.. autoclass:: SyncBatchNorm3d .. autoclass:: BatchNorm3d
:members: :members:
:hidden:`Inspiration` :hidden:`Inspiration`
...@@ -44,12 +44,6 @@ Customized NN modules in Encoding Package. For Synchronized Cross-GPU Batch Norm ...@@ -44,12 +44,6 @@ Customized NN modules in Encoding Package. For Synchronized Cross-GPU Batch Norm
.. autoclass:: UpsampleConv2d .. autoclass:: UpsampleConv2d
:members: :members:
:hidden:`DilatedAvgPool2d`
~~~~~~~~~~~~~~~~~~~~~~~~~~
.. autoclass:: DilatedAvgPool2d
:members:
:hidden:`GramMatrix` :hidden:`GramMatrix`
~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~
......
docs/source/syncbn.rst deleted 100644 → 0
View file @ 7e19143c
.. role:: hidden
:class: hidden-section
Synchronized BatchNorm
======================
.. note::
The original ``Self-Parallel`` version of ``BatchNorm`` has been deprecated in favor of PyTorch Compatible :class:`encoding.nn.BatchNorm2d`.
.. currentmodule:: encoding.nn
Modules
-------
:hidden:`BatchNorm1d`
~~~~~~~~~~~~~~~~~~~~~~~~
.. autoclass:: BatchNorm1d
:members:
:hidden:`BatchNorm2d`
~~~~~~~~~~~~~~~~~~~~~~~~
.. autoclass:: BatchNorm2d
:members:
.. currentmodule:: encoding
Functions
---------
.. currentmodule:: encoding.functions
:hidden:`batchnormtrain`
~~~~~~~~~~~~~~~~~~~~~~~~
.. autofunction:: batchnormtrain
:hidden:`batchnormeval`
~~~~~~~~~~~~~~~~~~~~~~~
.. autofunction:: batchnormeval
docs/source/utils.rst
View file @ 2d21747a
...@@ -9,10 +9,11 @@ Useful util functions. ...@@ -9,10 +9,11 @@ Useful util functions.
.. automodule:: encoding.utils .. automodule:: encoding.utils
.. currentmodule:: encoding.utils .. currentmodule:: encoding.utils
:hidden:`get_optimizer` :hidden:`LR_Scheduler`
~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~
.. autofunction:: get_optimizer .. autoclass:: LR_Scheduler
:members:
:hidden:`save_checkpoint` :hidden:`save_checkpoint`
~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~~
...@@ -20,11 +21,11 @@ Useful util functions. ...@@ -20,11 +21,11 @@ Useful util functions.
.. autofunction:: save_checkpoint .. autofunction:: save_checkpoint
:hidden:`batch_pix_accuracy` :hidden:`batch_pix_accuracy`
~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
.. autofunction:: batch_pix_accuracy .. autofunction:: batch_pix_accuracy
:hidden:`batch_intersection_union` :hidden:`batch_intersection_union`
~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
.. autofunction:: batch_intersection_union .. autofunction:: batch_intersection_union
encoding/CMakeLists.txt deleted 100644 → 0
View file @ 7e19143c
##+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
## Created by: Hang Zhang
## ECE Department, Rutgers University
## Email: zhang.hang@rutgers.edu
## Copyright (c) 2017
##
## This source code is licensed under the MIT-style license found in the
## LICENSE file in the root directory of this source tree
##+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
CMAKE_MINIMUM_REQUIRED(VERSION 2.8 FATAL_ERROR)
CMAKE_POLICY(VERSION 2.8)
INCLUDE(${CMAKE_CURRENT_SOURCE_DIR}/cmake/FindTorch.cmake)
IF(NOT CUDA_FOUND)
FIND_PACKAGE(CUDA 6.5 REQUIRED)
ENDIF()
# Detect CUDA architecture and get best NVCC flags
IF(NOT COMMAND CUDA_SELECT_NVCC_ARCH_FLAGS OR MSVC)
INCLUDE(${CMAKE_CURRENT_SOURCE_DIR}/cmake/select_compute_arch.cmake)
ENDIF()
LIST(APPEND CUDA_NVCC_FLAGS $ENV{TORCH_NVCC_FLAGS})
CUDA_SELECT_NVCC_ARCH_FLAGS(NVCC_FLAGS_EXTRA $ENV{TORCH_CUDA_ARCH_LIST})
LIST(APPEND CUDA_NVCC_FLAGS ${NVCC_FLAGS_EXTRA})
if(CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
if(CMAKE_CXX_COMPILER_VERSION VERSION_GREATER "4.9.3")
if(CUDA_VERSION VERSION_LESS "8.0")
MESSAGE(STATUS "Found gcc >=5 and CUDA <= 7.5, adding workaround C++ flags")
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -D_FORCE_INLINES -D_MWAITXINTRIN_H_INCLUDED -D__STRICT_ANSI__")
endif(CUDA_VERSION VERSION_LESS "8.0")
endif(CMAKE_CXX_COMPILER_VERSION VERSION_GREATER "4.9.3")
endif(CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
IF(MSVC)
LIST(APPEND CUDA_NVCC_FLAGS "-Xcompiler /wd4819")
ADD_DEFINITIONS(-DTH_EXPORTS)
ENDIF()
IF(NOT ENCODING_INSTALL_LIB_SUBDIR)
SET(ENCODING_INSTALL_LIB_SUBDIR "${TORCH_BUILD_DIR}" CACHE PATH "ENCODING install library directory")
SET(ENCODING_INSTALL_INCLUDE_SUBDIR "${TORCH_BUILD_DIR}/include" CACHE PATH "ENCODING install include subdirectory")
ENDIF()
SET(CMAKE_MACOSX_RPATH 1)
SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11")
FILE(GLOB src-cuda kernel/*.cu)
MESSAGE(STATUS "Torch_INSTALL_INCLUDE:" ${Torch_INSTALL_INCLUDE})
CUDA_INCLUDE_DIRECTORIES(
${CMAKE_CURRENT_SOURCE_DIR}/kernel
${CMAKE_CURRENT_SOURCE_DIR}/kernel/include
${Torch_INSTALL_INCLUDE}
)
CUDA_ADD_LIBRARY(ENCODING SHARED ${src-cuda})
IF(MSVC)
SET_TARGET_PROPERTIES(ENCODING PROPERTIES PREFIX "lib" IMPORT_PREFIX "lib")
ENDIF()
TARGET_LINK_LIBRARIES(ENCODING
${TH_LIBRARIES}
${CUDA_cusparse_LIBRARY}
)
# Luarocks bug pre-14.04 prevents us from setting it for Lua-Torch
IF(ENCODING_SO_VERSION)
MESSAGE(STATUS "ENCODING_SO_VERSION: ${ENCODING_SO_VERSION}")
SET_TARGET_PROPERTIES(ENCODING PROPERTIES
VERSION ${ENCODING_SO_VERSION}
SOVERSION ${ENCODING_SO_VERSION})
ENDIF(ENCODING_SO_VERSION)
#INSTALL(TARGETS ENCODING LIBRARY DESTINATION ${ENCODING_INSTALL_LIB_SUBDIR})
encoding/__init__.py
View file @ 2d21747a
...@@ -10,4 +10,4 @@ ...@@ -10,4 +10,4 @@
"""An optimized PyTorch package with CUDA backend.""" """An optimized PyTorch package with CUDA backend."""
from .version import __version__ from .version import __version__
from . import nn, functions, dilated, parallel, utils from . import nn, functions, dilated, parallel, utils, models, datasets
encoding/cmake/FindTorch.cmake deleted 100644 → 0
View file @ 7e19143c
##+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
## Created by: Hang Zhang
## ECE Department, Rutgers University
## Email: zhang.hang@rutgers.edu
## Copyright (c) 2017
##
## This source code is licensed under the MIT-style license found in the
## LICENSE file in the root directory of this source tree
##+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
# No longer using manual way to find the library.
if(FALSE)
FILE(GLOB TORCH_LIB_HINTS
"/anaconda/lib/python3.6/site-packages/torch/lib"
"/anaconda2/lib/python3.6/site-packages/torch/lib"
"$ENV{HOME}/anaconda/lib/python2.7/site-packages/torch/lib"
"$ENV{HOME}/anaconda2/lib/python2.7/site-packages/torch/lib"
)
FIND_PATH(TORCH_BUILD_DIR
NAMES "THNN.h"
PATHS "${TORCH_LIB_HINTS}"
)
FIND_LIBRARY(THC_LIBRARIES NAMES THC THC.1 PATHS ${TORCH_BUILD_DIR} PATH_SUFFIXES lib)
FIND_LIBRARY(TH_LIBRARIES NAMES TH TH.1 PATHS ${TORCH_BUILD_DIR} PATH_SUFFIXES lib)
endif()
# Set the envrionment variable via python
SET(TORCH_BUILD_DIR "$ENV{TORCH_BUILD_DIR}")
MESSAGE(STATUS "TORCH_BUILD_DIR: " ${TORCH_BUILD_DIR})
# Find the include files
SET(TORCH_TH_INCLUDE_DIR "${TORCH_BUILD_DIR}/include/TH")
SET(TORCH_THC_INCLUDE_DIR "${TORCH_BUILD_DIR}/include/THC")
SET(Torch_INSTALL_INCLUDE "${TORCH_BUILD_DIR}/include" ${TORCH_TH_INCLUDE_DIR} ${TORCH_THC_INCLUDE_DIR})
# Find the libs. We need to find libraries one by one.
SET(TH_LIBRARIES "$ENV{TH_LIBRARIES}")
encoding/cmake/select_compute_arch.cmake deleted 100644 → 0
View file @ 7e19143c
# Synopsis:
# CUDA_SELECT_NVCC_ARCH_FLAGS(out_variable [target_CUDA_architectures])
# -- Selects GPU arch flags for nvcc based on target_CUDA_architectures
# target_CUDA_architectures : Auto | Common | All | LIST(ARCH_AND_PTX ...)
# - "Auto" detects local machine GPU compute arch at runtime.
# - "Common" and "All" cover common and entire subsets of architectures
# ARCH_AND_PTX : NAME | NUM.NUM | NUM.NUM(NUM.NUM) | NUM.NUM+PTX
# NAME: Fermi Kepler Maxwell Kepler+Tegra Kepler+Tesla Maxwell+Tegra Pascal
# NUM: Any number. Only those pairs are currently accepted by NVCC though:
# 2.0 2.1 3.0 3.2 3.5 3.7 5.0 5.2 5.3 6.0 6.2
# Returns LIST of flags to be added to CUDA_NVCC_FLAGS in ${out_variable}
# Additionally, sets ${out_variable}_readable to the resulting numeric list
# Example:
# CUDA_SELECT_NVCC_ARCH_FLAGS(ARCH_FLAGS 3.0 3.5+PTX 5.2(5.0) Maxwell)
# LIST(APPEND CUDA_NVCC_FLAGS ${ARCH_FLAGS})
#
# More info on CUDA architectures: https://en.wikipedia.org/wiki/CUDA
#
# This list will be used for CUDA_ARCH_NAME = All option
set(CUDA_KNOWN_GPU_ARCHITECTURES "Fermi" "Kepler" "Maxwell")
# This list will be used for CUDA_ARCH_NAME = Common option (enabled by default)
set(CUDA_COMMON_GPU_ARCHITECTURES "3.0" "3.5" "5.0")
if (CUDA_VERSION VERSION_GREATER "6.5")
list(APPEND CUDA_KNOWN_GPU_ARCHITECTURES "Kepler+Tegra" "Kepler+Tesla" "Maxwell+Tegra")
list(APPEND CUDA_COMMON_GPU_ARCHITECTURES "5.2")
endif ()
if (CUDA_VERSION VERSION_GREATER "7.5")
list(APPEND CUDA_KNOWN_GPU_ARCHITECTURES "Pascal")
list(APPEND CUDA_COMMON_GPU_ARCHITECTURES "6.0" "6.1" "6.1+PTX")
else()
list(APPEND CUDA_COMMON_GPU_ARCHITECTURES "5.2+PTX")
endif ()
################################################################################################
# A function for automatic detection of GPUs installed (if autodetection is enabled)
# Usage:
# CUDA_DETECT_INSTALLED_GPUS(OUT_VARIABLE)
#
function(CUDA_DETECT_INSTALLED_GPUS OUT_VARIABLE)
if(NOT CUDA_GPU_DETECT_OUTPUT)
set(cufile ${PROJECT_BINARY_DIR}/detect_cuda_archs.cu)
file(WRITE ${cufile} ""
"#include <cstdio>\n"
"int main()\n"
"{\n"
" int count = 0;\n"
" if (cudaSuccess != cudaGetDeviceCount(&count)) return -1;\n"
" if (count == 0) return -1;\n"
" for (int device = 0; device < count; ++device)\n"
" {\n"
" cudaDeviceProp prop;\n"
" if (cudaSuccess == cudaGetDeviceProperties(&prop, device))\n"
" std::printf(\"%d.%d \", prop.major, prop.minor);\n"
" }\n"
" return 0;\n"
"}\n")
execute_process(COMMAND "${CUDA_NVCC_EXECUTABLE}" "--run" "${cufile}"
"-ccbin" ${CMAKE_CXX_COMPILER}
WORKING_DIRECTORY "${PROJECT_BINARY_DIR}/CMakeFiles/"
RESULT_VARIABLE nvcc_res OUTPUT_VARIABLE nvcc_out
ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
if(nvcc_res EQUAL 0)
# only keep the last line of nvcc_out
string(REGEX REPLACE ";" "\\\\;" nvcc_out "${nvcc_out}")
string(REGEX REPLACE "\n" ";" nvcc_out "${nvcc_out}")
list(GET nvcc_out -1 nvcc_out)
string(REPLACE "2.1" "2.1(2.0)" nvcc_out "${nvcc_out}")
set(CUDA_GPU_DETECT_OUTPUT ${nvcc_out} CACHE INTERNAL "Returned GPU architetures from detect_gpus tool" FORCE)
endif()
endif()
if(NOT CUDA_GPU_DETECT_OUTPUT)
message(STATUS "Automatic GPU detection failed. Building for common architectures.")
set(${OUT_VARIABLE} ${CUDA_COMMON_GPU_ARCHITECTURES} PARENT_SCOPE)
else()
set(${OUT_VARIABLE} ${CUDA_GPU_DETECT_OUTPUT} PARENT_SCOPE)
endif()
endfunction()
################################################################################################
# Function for selecting GPU arch flags for nvcc based on CUDA architectures from parameter list
# Usage:
# SELECT_NVCC_ARCH_FLAGS(out_variable [list of CUDA compute archs])
function(CUDA_SELECT_NVCC_ARCH_FLAGS out_variable)
set(CUDA_ARCH_LIST "${ARGN}")
if("X${CUDA_ARCH_LIST}" STREQUAL "X" )
set(CUDA_ARCH_LIST "Auto")
endif()
set(cuda_arch_bin)
set(cuda_arch_ptx)
if("${CUDA_ARCH_LIST}" STREQUAL "All")
set(CUDA_ARCH_LIST ${CUDA_KNOWN_GPU_ARCHITECTURES})
elseif("${CUDA_ARCH_LIST}" STREQUAL "Common")
set(CUDA_ARCH_LIST ${CUDA_COMMON_GPU_ARCHITECTURES})
elseif("${CUDA_ARCH_LIST}" STREQUAL "Auto")
CUDA_DETECT_INSTALLED_GPUS(CUDA_ARCH_LIST)
message(STATUS "Autodetected CUDA architecture(s): ${CUDA_ARCH_LIST}")
endif()
# Now process the list and look for names
string(REGEX REPLACE "[ \t]+" ";" CUDA_ARCH_LIST "${CUDA_ARCH_LIST}")
list(REMOVE_DUPLICATES CUDA_ARCH_LIST)
foreach(arch_name ${CUDA_ARCH_LIST})
set(arch_bin)
set(add_ptx FALSE)
# Check to see if we are compiling PTX
if(arch_name MATCHES "(.*)\\+PTX$")
set(add_ptx TRUE)
set(arch_name ${CMAKE_MATCH_1})
endif()
if(arch_name MATCHES "(^[0-9]\\.[0-9](\\([0-9]\\.[0-9]\\))?)$")
set(arch_bin ${CMAKE_MATCH_1})
set(arch_ptx ${arch_bin})
else()
# Look for it in our list of known architectures
if(${arch_name} STREQUAL "Fermi")
set(arch_bin "2.0 2.1(2.0)")
elseif(${arch_name} STREQUAL "Kepler+Tegra")
set(arch_bin 3.2)
elseif(${arch_name} STREQUAL "Kepler+Tesla")
set(arch_bin 3.7)
elseif(${arch_name} STREQUAL "Kepler")
set(arch_bin 3.0 3.5)
set(arch_ptx 3.5)
elseif(${arch_name} STREQUAL "Maxwell+Tegra")
set(arch_bin 5.3)
elseif(${arch_name} STREQUAL "Maxwell")
set(arch_bin 5.0 5.2)
set(arch_ptx 5.2)
elseif(${arch_name} STREQUAL "Pascal")
set(arch_bin 6.0 6.1)
set(arch_ptx 6.1)
else()
message(SEND_ERROR "Unknown CUDA Architecture Name ${arch_name} in CUDA_SELECT_NVCC_ARCH_FLAGS")
endif()
endif()
if(NOT arch_bin)
message(SEND_ERROR "arch_bin wasn't set for some reason")
endif()
list(APPEND cuda_arch_bin ${arch_bin})
if(add_ptx)
if (NOT arch_ptx)
set(arch_ptx ${arch_bin})
endif()
list(APPEND cuda_arch_ptx ${arch_ptx})
endif()
endforeach()
# remove dots and convert to lists
string(REGEX REPLACE "\\." "" cuda_arch_bin "${cuda_arch_bin}")
string(REGEX REPLACE "\\." "" cuda_arch_ptx "${cuda_arch_ptx}")
string(REGEX MATCHALL "[0-9()]+" cuda_arch_bin "${cuda_arch_bin}")
string(REGEX MATCHALL "[0-9]+" cuda_arch_ptx "${cuda_arch_ptx}")
if(cuda_arch_bin)
list(REMOVE_DUPLICATES cuda_arch_bin)
endif()
if(cuda_arch_ptx)
list(REMOVE_DUPLICATES cuda_arch_ptx)
endif()
set(nvcc_flags "")
set(nvcc_archs_readable "")
# Tell NVCC to add binaries for the specified GPUs
foreach(arch ${cuda_arch_bin})
if(arch MATCHES "([0-9]+)\\(([0-9]+)\\)")
# User explicitly specified ARCH for the concrete CODE
list(APPEND nvcc_flags -gencode arch=compute_${CMAKE_MATCH_2},code=sm_${CMAKE_MATCH_1})
list(APPEND nvcc_archs_readable sm_${CMAKE_MATCH_1})
else()
# User didn't explicitly specify ARCH for the concrete CODE, we assume ARCH=CODE
list(APPEND nvcc_flags -gencode arch=compute_${arch},code=sm_${arch})
list(APPEND nvcc_archs_readable sm_${arch})
endif()
endforeach()
# Tell NVCC to add PTX intermediate code for the specified architectures
foreach(arch ${cuda_arch_ptx})
list(APPEND nvcc_flags -gencode arch=compute_${arch},code=compute_${arch})
list(APPEND nvcc_archs_readable compute_${arch})
endforeach()
string(REPLACE ";" " " nvcc_archs_readable "${nvcc_archs_readable}")
set(${out_variable} ${nvcc_flags} PARENT_SCOPE)
set(${out_variable}_readable ${nvcc_archs_readable} PARENT_SCOPE)
endfunction()
encoding/datasets/__init__.py 0 → 100644
View file @ 2d21747a
from .base import *
from .ade20k import ADE20KSegmentation
from .pascal_voc import VOCSegmentation
from .pascal_aug import VOCAugSegmentation
from .pcontext import ContextSegmentation
datasets = {
'ade20k': ADE20KSegmentation,
'pascal_voc': VOCSegmentation,
'pascal_aug': VOCAugSegmentation,
'pcontext': ContextSegmentation,
}
def get_segmentation_dataset(name, **kwargs):
return datasets[name.lower()](**kwargs)
encoding/datasets/ade20k.py 0 → 100644
View file @ 2d21747a
###########################################################################
# Created by: Hang Zhang
# Email: zhang.hang@rutgers.edu
# Copyright (c) 2017
###########################################################################
import os
import sys
import numpy as np
import random
import math
from PIL import Image, ImageOps, ImageFilter
import torch
import torch.utils.data as data
import torchvision.transforms as transform
from .base import BaseDataset
class ADE20KSegmentation(BaseDataset):
BASE_DIR = 'ADEChallengeData2016'
NUM_CLASS = 150
def __init__(self, root=os.path.expanduser('~/.encoding/data'), split='train',
mode=None, transform=None, target_transform=None):
super(ADE20KSegmentation, self).__init__(
root, split, mode, transform, target_transform)
# assert exists and prepare dataset automatically
root = os.path.join(root, self.BASE_DIR)
assert os.path.exists(root), "Please setup the dataset using" + \
"encoding/scripts/prepare_ade20k.py"
self.images, self.masks = _get_ade20k_pairs(root, split)
if split != 'test':
assert (len(self.images) == len(self.masks))
if len(self.images) == 0:
raise(RuntimeError("Found 0 images in subfolders of: \
" + root + "\n"))
def __getitem__(self, index):
img = Image.open(self.images[index]).convert('RGB')
if self.mode == 'test':
if self.transform is not None:
img = self.transform(img)
return img, os.path.basename(self.images[index])
mask = Image.open(self.masks[index])
# synchrosized transform
if self.mode == 'train':
img, mask = self._sync_transform(img, mask)
elif self.mode == 'val':
img, mask = self._val_sync_transform(img, mask)
else:
assert self.mode == 'testval'
mask = self._mask_transform(mask)
# general resize, normalize and toTensor
if self.transform is not None:
img = self.transform(img)
if self.target_transform is not None:
mask = self.target_transform(mask)
return img, mask
def _mask_transform(self, mask):
target = np.array(mask).astype('int32') - 1
return torch.from_numpy(target).long()
def __len__(self):
return len(self.images)
@property
def pred_offset(self):
return 1
def _get_ade20k_pairs(folder, split='train'):
img_paths = []
mask_paths = []
if split == 'train':
img_folder = os.path.join(folder, 'images/training')
mask_folder = os.path.join(folder, 'annotations/training')
else:
img_folder = os.path.join(folder, 'images/validation')
mask_folder = os.path.join(folder, 'annotations/validation')
for filename in os.listdir(img_folder):
basename, _ = os.path.splitext(filename)
if filename.endswith(".jpg"):
imgpath = os.path.join(img_folder, filename)
maskname = basename + '.png'
maskpath = os.path.join(mask_folder, maskname)
if os.path.isfile(maskpath):
img_paths.append(imgpath)
mask_paths.append(maskpath)
else:
print('cannot find the mask:', maskpath)
return img_paths, mask_paths
encoding/datasets/base.py 0 → 100644
View file @ 2d21747a
###########################################################################
# Created by: Hang Zhang
# Email: zhang.hang@rutgers.edu
# Copyright (c) 2017
###########################################################################
import random
import numpy as np
from PIL import Image, ImageOps, ImageFilter
import torch
import torch.utils.data as data
__all__ = ['BaseDataset', 'test_batchify_fn']
class BaseDataset(data.Dataset):
def __init__(self, root, split, mode=None, transform=None,
target_transform=None, base_size=520, crop_size=480):
self.root = root
self.transform = transform
self.target_transform = target_transform
self.split = split
self.mode = mode if mode is not None else split
self.base_size = base_size
self.crop_size = crop_size
def __getitem__(self, index):
raise NotImplemented
@property
def num_class(self):
return self.NUM_CLASS
@property
def pred_offset(self):
raise NotImplemented
def _val_sync_transform(self, img, mask):
outsize = self.crop_size
short_size = outsize
w, h = img.size
if w > h:
oh = short_size
ow = int(1.0 * w * oh / h)
else:
ow = short_size
oh = int(1.0 * h * ow / w)
img = img.resize((ow, oh), Image.BILINEAR)
mask = mask.resize((ow, oh), Image.NEAREST)
# center crop
w, h = img.size
x1 = int(round((w - outsize) / 2.))
y1 = int(round((h - outsize) / 2.))
img = img.crop((x1, y1, x1+outsize, y1+outsize))
mask = mask.crop((x1, y1, x1+outsize, y1+outsize))
# final transform
return img, self._mask_transform(mask)
def _sync_transform(self, img, mask):
# random mirror
if random.random() < 0.5:
img = img.transpose(Image.FLIP_LEFT_RIGHT)
mask = mask.transpose(Image.FLIP_LEFT_RIGHT)
crop_size = self.crop_size
# random scale (short edge from 480 to 720)
short_size = random.randint(int(self.base_size*0.5), int(self.base_size*2.0))
w, h = img.size
if h > w:
ow = short_size
oh = int(1.0 * h * ow / w)
else:
oh = short_size
ow = int(1.0 * w * oh / h)
img = img.resize((ow, oh), Image.BILINEAR)
mask = mask.resize((ow, oh), Image.NEAREST)
# random rotate -10~10, mask using NN rotate
deg = random.uniform(-10, 10)
img = img.rotate(deg, resample=Image.BILINEAR)
mask = mask.rotate(deg, resample=Image.NEAREST)
# pad crop
if short_size < crop_size:
padh = crop_size - oh if oh < crop_size else 0
padw = crop_size - ow if ow < crop_size else 0
img = ImageOps.expand(img, border=(0, 0, padw, padh), fill=0)
mask = ImageOps.expand(mask, border=(0, 0, padw, padh), fill=0)
# random crop crop_size
w, h = img.size
x1 = random.randint(0, w - crop_size)
y1 = random.randint(0, h - crop_size)
img = img.crop((x1, y1, x1+crop_size, y1+crop_size))
mask = mask.crop((x1, y1, x1+crop_size, y1+crop_size))
# gaussian blur as in PSP
if random.random() < 0.5:
img = img.filter(ImageFilter.GaussianBlur(
radius=random.random()))
# final transform
return img, self._mask_transform(mask)
def _mask_transform(self, mask):
return torch.from_numpy(np.array(mask)).long()
def test_batchify_fn(data):
error_msg = "batch must contain tensors, tuples or lists; found {}"
if isinstance(data[0], (str, torch.Tensor)):
return list(data)
elif isinstance(data[0], (tuple, list)):
data = zip(*data)
return [test_batchify_fn(i) for i in data]
elif isinstance(data[0], ):
data = np.asarray(data)
return mx.nd.array(data, dtype=data.dtype)
raise TypeError((error_msg.format(type(batch[0]))))
encoding/datasets/cityscapes.py 0 → 100644
View file @ 2d21747a
###########################################################################
# Created by: Hang Zhang
# Email: zhang.hang@rutgers.edu
# Copyright (c) 2017
###########################################################################
import os
import sys
import numpy as np
import random
import math
from tqdm import tqdm
from PIL import Image, ImageOps, ImageFilter
import torch
import torch.utils.data as data
import torchvision.transforms as transform
class Segmentation(data.Dataset):
BASE_DIR = 'cityscapes'
def __init__(self, data_folder, mode='train', transform=None,
target_transform=None):
self.root = os.path.join(data_folder, self.BASE_DIR)
self.transform = transform
self.target_transform = target_transform
self.mode = mode
self.images, self.masks = get_city_pairs(self.root, mode)
assert (len(self.images) == len(self.masks))
print("Found {} images in subfolders of: {}\
".format(len(self.images), self.root + '/' + mode))
if len(self.images) == 0:
raise(RuntimeError("Found 0 images in subfolders of: \
" + self.root + "\n"))
def __getitem__(self, index):
img = Image.open(self.images[index]).convert('RGB')
if self.mode == 'test':
if self.transform is not None:
img = self.transform(img)
return img, os.path.basename(self.images[index])
mask = Image.open(self.masks[index])#.convert("P")
mask = np.array(mask)
mask += 1
mask[mask==256] = 0
mask = Image.fromarray(mask)
# synchrosized transform
if self.mode == 'train':
img, mask = self._sync_transform(img, mask)
elif self.mode == 'val':
img, mask = self._val_sync_transform(img, mask)
else:
raise RuntimeError('unknown mode for dataloader: {}'.format(self.mode))
# general resize, normalize and toTensor
if self.transform is not None:
#print("transform for input")
img = self.transform(img)
if self.target_transform is not None:
#print("transform for label")
mask = self.target_transform(mask)
return img, mask
def __len__(self):
return len(self.images)
def _val_sync_transform(self, img, mask):
"""
synchronized transformation
"""
outsize = 720
short = outsize
w, h = img.size
if w > h:
oh = short
ow = int(1.0 * w * oh / h)
else:
ow = short
oh = int(1.0 * h * ow / w)
img = img.resize((ow, oh), Image.BILINEAR)
mask = mask.resize((ow, oh), Image.NEAREST)
# center crop
w, h = img.size
x1 = int(round((w - outsize) / 2.))
y1 = int(round((h - outsize) / 2.))
img = img.crop((x1, y1, x1+outsize, y1+outsize))
mask = mask.crop((x1, y1, x1+outsize, y1+outsize))
return img, mask
def _sync_transform(self, img, mask):
# random mirror
if random.random() < 0.5:
img = img.transpose(Image.FLIP_LEFT_RIGHT)
mask = mask.transpose(Image.FLIP_LEFT_RIGHT)
base_size = 2048
crop_size = 720
# random scale (short edge from 480 to 720)
long_size = random.randint(int(base_size*0.5), int(base_size*2.0))
w, h = img.size
if h > w:
oh = long_size
ow = int(1.0 * w * oh / h)
short_size = ow
else:
ow = long_size
oh = int(1.0 * h * ow / w)
short_size = oh
img = img.resize((ow, oh), Image.BILINEAR)
mask = mask.resize((ow, oh), Image.NEAREST)
# random rotate -10~10, mask using NN rotate
deg = random.uniform(-10,10)
img = img.rotate(deg, resample=Image.BILINEAR)
mask = mask.rotate(deg, resample=Image.NEAREST)
# pad crop
if short_size < crop_size:
padh = crop_size - oh if oh < crop_size else 0
padw = crop_size - ow if ow < crop_size else 0
img = ImageOps.expand(img, border=(0,0,padw,padh), fill=0)
mask = ImageOps.expand(mask, border=(0,0,padw,padh), fill=0)
# random crop 480
w, h = img.size
x1 = random.randint(0, w - crop_size)
y1 = random.randint(0, h - crop_size)
img = img.crop((x1, y1, x1+crop_size, y1+crop_size))
mask = mask.crop((x1, y1, x1+crop_size, y1+crop_size))
# gaussian blur as in PSP ?
if random.random() < 0.5:
img = img.filter(ImageFilter.GaussianBlur(
radius=random.random()))
return img, mask
def get_city_pairs(folder, mode='train'):
img_paths = []
mask_paths = []
img_folder = os.path.join(folder, 'leftImg8bit/' + mode)
mask_folder = os.path.join(folder, 'gtFine/'+ mode)
for root, directories, files in os.walk(img_folder):
for filename in files:#os.listdir(img_folder):
#basename, extension =os.path.splitext(filename)
if filename.endswith(".png"):
imgpath = os.path.join(root, filename)
foldername = os.path.basename(os.path.dirname(imgpath))
maskname = filename.replace('leftImg8bit','gtFine_trainIds')
maskpath = os.path.join(mask_folder, foldername, maskname)
if os.path.isfile(imgpath) and os.path.isfile(maskpath):
img_paths.append(imgpath)
mask_paths.append(maskpath)
else:
print('cannot find the mask or image:', imgpath, maskpath)
return img_paths, mask_paths
encoding/datasets/cityscapescoarse.py 0 → 100644
View file @ 2d21747a
###########################################################################
# Created by: Hang Zhang
# Email: zhang.hang@rutgers.edu
# Copyright (c) 2017
###########################################################################
import os
import sys
import numpy as np
import random
import math
from tqdm import tqdm
from PIL import Image, ImageOps, ImageFilter
import torch
import torch.utils.data as data
import torchvision.transforms as transform
class Segmentation(data.Dataset):
BASE_DIR = 'cityscapes'
def __init__(self, data_folder, mode='train', transform=None,
target_transform=None):
self.root = os.path.join(data_folder, self.BASE_DIR)
self.transform = transform
self.target_transform = target_transform
self.mode = mode
self.images, self.masks = get_city_pairs(self.root, mode)
assert (len(self.images) == len(self.masks))
if len(self.images) == 0:
raise(RuntimeError("Found 0 images in subfolders of: \
" + self.root + "\n"))
def __getitem__(self, index):
img = Image.open(self.images[index]).convert('RGB')
if self.mode == 'test':
if self.transform is not None:
img = self.transform(img)
return img, os.path.basename(self.images[index])
mask = Image.open(self.masks[index])#.convert("P")
mask = np.array(mask)
mask += 1
mask[mask==256] = 0
mask = Image.fromarray(mask)
# synchrosized transform
if self.mode == 'train':
img, mask = self._sync_transform(img, mask)
elif self.mode == 'val':
img, mask = self._val_sync_transform(img, mask)
else:
raise RuntimeError('unknown mode for dataloader: {}'.format(self.mode))
# general resize, normalize and toTensor
if self.transform is not None:
#print("transform for input")
img = self.transform(img)
if self.target_transform is not None:
#print("transform for label")
mask = self.target_transform(mask)
return img, mask
def __len__(self):
return len(self.images)
def _val_sync_transform(self, img, mask):
"""
synchronized transformation
"""
outsize = 720
short = outsize
w, h = img.size
if w > h:
oh = short
ow = int(1.0 * w * oh / h)
else:
ow = short
oh = int(1.0 * h * ow / w)
img = img.resize((ow, oh), Image.BILINEAR)
mask = mask.resize((ow, oh), Image.NEAREST)
# center crop
w, h = img.size
x1 = int(round((w - outsize) / 2.))
y1 = int(round((h - outsize) / 2.))
img = img.crop((x1, y1, x1+outsize, y1+outsize))
mask = mask.crop((x1, y1, x1+outsize, y1+outsize))
return img, mask
def _sync_transform(self, img, mask):
# random mirror
if random.random() < 0.5:
img = img.transpose(Image.FLIP_LEFT_RIGHT)
mask = mask.transpose(Image.FLIP_LEFT_RIGHT)
base_size = 2048
crop_size = 720
# random scale (short edge from 480 to 720)
long_size = random.randint(int(base_size*0.5), int(base_size*2.0))
w, h = img.size
if h > w:
oh = long_size
ow = int(1.0 * w * oh / h)
short_size = ow
else:
ow = long_size
oh = int(1.0 * h * ow / w)
short_size = oh
img = img.resize((ow, oh), Image.BILINEAR)
mask = mask.resize((ow, oh), Image.NEAREST)
# random rotate -10~10, mask using NN rotate
deg = random.uniform(-10,10)
img = img.rotate(deg, resample=Image.BILINEAR)
mask = mask.rotate(deg, resample=Image.NEAREST)
# pad crop
if short_size < crop_size:
padh = crop_size - oh if oh < crop_size else 0
padw = crop_size - ow if ow < crop_size else 0
img = ImageOps.expand(img, border=(0,0,padw,padh), fill=0)
mask = ImageOps.expand(mask, border=(0,0,padw,padh), fill=0)
# random crop 480
w, h = img.size
x1 = random.randint(0, w - crop_size)
y1 = random.randint(0, h - crop_size)
img = img.crop((x1, y1, x1+crop_size, y1+crop_size))
mask = mask.crop((x1, y1, x1+crop_size, y1+crop_size))
# gaussian blur as in PSP ?
if random.random() < 0.5:
img = img.filter(ImageFilter.GaussianBlur(
radius=random.random()))
return img, mask
def get_city_pairs(folder, mode='train'):
img_paths = []
mask_paths = []
if mode=='train':
img_folder = os.path.join(folder, 'leftImg8bit/train_extra')
mask_folder = os.path.join(folder, 'gtCoarse/train_extra')
else:
img_folder = os.path.join(folder, 'leftImg8bit/val')
mask_folder = os.path.join(folder, 'gtFine/val')
for root, directories, files in os.walk(img_folder):
for filename in files:
basename, extension =os.path.splitext(filename)
if filename.endswith(".png"):
imgpath = os.path.join(root, filename)
foldername = os.path.basename(os.path.dirname(imgpath))
maskname = filename.replace('leftImg8bit','gtCoarse_trainIds')
maskpath = os.path.join(mask_folder, foldername, maskname)
if os.path.isfile(imgpath) and os.path.isfile(maskpath):
img_paths.append(imgpath)
mask_paths.append(maskpath)
else:
print('cannot find the mask or image:', imgpath, maskpath)
return img_paths, mask_paths
encoding/datasets/coco.py 0 → 100644
View file @ 2d21747a
import os
from tqdm import tqdm, trange
from PIL import Image, ImageOps, ImageFilter
import torch
import torch.utils.data as data
import torchvision.transforms as transform
import random
import math
import numpy as np
from .dataset import ToLabel
"""
NUM_CHANNEL = 91
[] background
[5] airplane
[2] bicycle
[16] bird
[9] boat
[44] bottle
[6] bus
[3] car
[17] cat
[62] chair
[21] cow
[67] dining table
[18] dog
[19] horse
[4] motorcycle
[1] person
[64] potted plant
[20] sheep
[63] couch
[7] train
[72] tv
"""
catlist = [0, 5, 2, 16, 9, 44, 6, 3, 17, 62, 21, 67, 18, 19, 4,
1, 64, 20, 63, 7, 72]
class Segmentation(data.Dataset):
def __init__(self, root, mode='train', transform=None,
target_transform=None):
from pycocotools.coco import COCO
from pycocotools import mask
if mode == 'train':
print('train set')
ann_file = os.path.join(root, 'coco/annotations/instances_train2014.json')
ids_file = os.path.join(root, 'coco/annotations/train_ids.pth')
root = os.path.join(root, 'coco/train2014')
else:
print('val set')
ann_file = os.path.join(root, 'coco/annotations/instances_val2014.json')
ids_file = os.path.join(root, 'coco/annotations/val_ids.pth')
root = os.path.join(root, 'coco/val2014')
self.train = mode
self.root = root
self.coco = COCO(ann_file)
self.coco_mask = mask
if os.path.exists(ids_file):
self.ids = torch.load(ids_file)
else:
self.new_ids = []
self.ids = list(self.coco.imgs.keys())
self.preprocess(ids_file)
self.ids = self.new_ids
self.transform = transform
self.target_transform = target_transform
def preprocess(self, ids_file):
tbar = trange(len(self.ids))
for i in tbar:
img_id = self.ids[i]
cocotarget = self.coco.loadAnns(self.coco.getAnnIds(imgIds=img_id))
img_metadata = self.coco.loadImgs(img_id)[0]
mask = self._gen_seg_mask(cocotarget, img_metadata['height'],
img_metadata['width'])
# more than 1k pixels
if (mask > 0).sum() > 1000:
self.new_ids.append(img_id)
tbar.set_description('Doing: {}/{}, got {} qualified images'.\
format(i, len(self.ids), len(self.new_ids)))
print('number of qualified images: ', len(self.new_ids))
torch.save(self.new_ids, ids_file)
def __getitem__(self, index):
coco = self.coco
img_id = self.ids[index]
cocotarget = coco.loadAnns(coco.getAnnIds(imgIds=img_id))
img_metadata = coco.loadImgs(img_id)[0]
path = img_metadata['file_name']
img = Image.open(os.path.join(self.root, path)).convert('RGB')
mask = Image.fromarray(
self._gen_seg_mask(cocotarget, img_metadata['height'],
img_metadata['width'])
)
# synchrosized transform
if True:#self.train == 'train':
img, mask = self._sync_transform(img, mask)
else:
img, mask = self._val_sync_transform(img, mask)
# general resize, normalize and toTensor
if self.transform is not None:
#print("transform for input")
img = self.transform(img)
if self.target_transform is not None:
#print("transform for label")
mask = self.target_transform(mask)
return img, mask
def __len__(self):
return len(self.ids)
def _val_sync_transform(self, img, mask):
outsize = 480
short = outsize
w, h = img.size
if w > h:
oh = short
ow = int(1.0 * w * oh / h)
else:
ow = short
oh = int(1.0 * h * ow / w)
img = img.resize((ow, oh), Image.BILINEAR)
mask = mask.resize((ow, oh), Image.NEAREST)
# center crop
w, h = img.size
x1 = int(round((w - outsize) / 2.))
y1 = int(round((h - outsize) / 2.))
img = img.crop((x1, y1, x1+outsize, y1+outsize))
mask = mask.crop((x1, y1, x1+outsize, y1+outsize))
return img, mask
def _sync_transform(self, img, mask):
# random mirror
if random.random() < 0.5:
img = img.transpose(Image.FLIP_LEFT_RIGHT)
mask = mask.transpose(Image.FLIP_LEFT_RIGHT)
base_size = 520
crop_size = 480
# random scale (short edge from 480 to 720)
long_size = random.randint(int(base_size*0.5), int(base_size*2.0))
w, h = img.size
if h > w:
oh = long_size
ow = int(1.0 * w * oh / h)
short_size = ow
else:
ow = long_size
oh = int(1.0 * h * ow / w)
short_size = oh
img = img.resize((ow, oh), Image.BILINEAR)
mask = mask.resize((ow, oh), Image.NEAREST)
# random rotate -10~10, mask using NN rotate
deg = random.uniform(-10,10)
img = img.rotate(deg, resample=Image.BILINEAR)
mask = mask.rotate(deg, resample=Image.NEAREST)
# pad crop
if short_size < crop_size:
padh = crop_size - oh if oh < crop_size else 0
padw = crop_size - ow if ow < crop_size else 0
img = ImageOps.expand(img, border=(0,0,padw,padh), fill=0)
mask = ImageOps.expand(mask, border=(0,0,padw,padh), fill=0)
# random crop 480
w, h = img.size
x1 = random.randint(0, w - crop_size)
y1 = random.randint(0, h - crop_size)
img = img.crop((x1, y1, x1+crop_size, y1+crop_size))
mask = mask.crop((x1, y1, x1+crop_size, y1+crop_size))
# gaussian blur as in PSP ?
if random.random() < 0.5:
img = img.filter(ImageFilter.GaussianBlur(
radius=random.random()))
return img, mask
def _gen_seg_mask(self, target, h, w):
mask = np.zeros((h, w), dtype=np.uint8)
coco_mask = self.coco_mask
for instance in target:
rle = coco_mask.frPyObjects(instance['segmentation'], h, w)
m = coco_mask.decode(rle)
cat = instance['category_id']
if cat in catlist:
c = catlist.index(cat)
else:
continue
if len(m.shape) < 3:
mask[:, :] += (mask == 0) * (m * c)
else:
mask[:, :] += (mask == 0) * (((np.sum(m, axis=2)) > 0) * c).astype(np.uint8)
return mask
encoding/datasets/download_datasets.sh 0 → 100644
View file @ 2d21747a
mkdir -p "$HOME"/data
cd "$HOME"/data
# augmented PASCAL VOC
wget http://www.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/semantic_contours/benchmark.tgz # 1.3 GB
tar -zxvf benchmark.tgz
mv benchmark_RELEASE VOCaug
<<<<<<< HEAD
# generate trainval txt
cd VOCaug/dataset/
cp train.txt trainval.txt
cat val.txt >> trainval.txt
cd -
=======
# generate trainval.txt
cd VOCaug/dataset/
cp train.txt trainval.txt
cat val.txt >> trainval.txt
cd "$HOME"/data
>>>>>>> d7e511b09ee6127be11f03497b7d83327e9f7b1b
# original PASCAL VOC 2012
wget http://host.robots.ox.ac.uk/pascal/VOC/voc2012/VOCtrainval_11-May-2012.tar # 2 GB
tar -xvf VOCtrainval_11-May-2012.tar
<<<<<<< HEAD
# for PASCAL VOC testset, you need to login and manually download from (http://host.robots.ox.ac.uk:8080/)
=======
cd -
>>>>>>> d7e511b09ee6127be11f03497b7d83327e9f7b1b
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment