def classifier(img_path, model_name):
# load the image
try:
img_pil = Image.open(img_path)
except: # has not been tested
sys.exit("Exception: unable to open " + img_path)
# define transforms
preprocess = transforms.Compose([
transforms.Resize(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])
# preprocess the image
img_tensor = preprocess(img_pil)
# resize the tensor (add dimension for batch)
img_tensor.unsqueeze_(0)
# wrap input in variable, wrap input in variable - no longer needed for
# v 0.4 & higher code changed 04/26/2018 by Jennifer S. to handle PyTorch upgrade
pytorch_ver = __version__.split('.')
# pytorch versions 0.4 & hihger - Variable depreciated so that it returns
# a tensor. So to address tensor as output (not wrapper) and to mimic the
# affect of setting volatile = True (because we are using pretrained models
# for inference) we can set requires_gradient to False. Here we just set
# requires_grad_ to False on our tensor
if int(pytorch_ver[0]) > 0 or int(pytorch_ver[1]) >= 4:
img_tensor.requires_grad_(False)
# pytorch versions less than 0.4 - uses Variable because not-depreciated
else:
# apply model to input
# wrap input in variable
data = Variable(img_tensor, volatile = True)
# apply model to input
model = models[model_name]
# puts model in evaluation mode
# instead of (default)training mode
model = model.eval()
# apply data to model - adjusted based upon version to account for
# operating on a Tensor for version 0.4 & higher.
if int(pytorch_ver[0]) > 0 or int(pytorch_ver[1]) >= 4:
output = model(img_tensor)
# pytorch versions less than 0.4
else:
# apply data to model
output = model(data)
# return index corresponding to predicted class
pred_idx = output.data.numpy().argmax()
return imagenet_classes_dict[pred_idx]
def classifier(img_path, model_name):
# load the image
img_pil = Image.open(img_path)
# define transforms
preprocess = transforms.Compose([
transforms.Resize(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])
# preprocess the image
img_tensor = preprocess(img_pil)
# resize the tensor (add dimension for batch)
img_tensor.unsqueeze_(0)
# wrap input in variable, wrap input in variable - no longer needed for
# v 0.4 & higher code changed 04/26/2018 by Jennifer S. to handle PyTorch upgrade
pytorch_ver = __version__.split('.')
# pytorch versions 0.4 & hihger - Variable depreciated so that it returns
# a tensor. So to address tensor as output (not wrapper) and to mimic the
# affect of setting volatile = True (because we are using pretrained models
# for inference) we can set requires_gradient to False. Here we just set
# requires_grad_ to False on our tensor
if int(pytorch_ver[0]) > 0 or int(pytorch_ver[1]) >= 4:
img_tensor.requires_grad_(False)
# pytorch versions less than 0.4 - uses Variable because not-depreciated
else:
# apply model to input
# wrap input in variable
data = Variable(img_tensor, volatile = True)
# apply model to input
model = models[model_name]
# puts model in evaluation mode
# instead of (default)training mode
model = model.eval()
# apply data to model - adjusted based upon version to account for
# operating on a Tensor for version 0.4 & higher.
if int(pytorch_ver[0]) > 0 or int(pytorch_ver[1]) >= 4:
output = model(img_tensor)
# pytorch versions less than 0.4
else:
# apply data to model
output = model(data)
# return index corresponding to predicted class
pred_idx = output.data.numpy().argmax()
return imagenet_classes_dict[pred_idx]
def classifier(img_path, model_name):
# load the image
img_pil = Image.open(img_path)
# define transforms
preprocess = transforms.Compose([
transforms.Resize(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])
# preprocess the image
img_tensor = preprocess(img_pil)
# resize the tensor (add dimension for batch)
img_tensor.unsqueeze_(0)
# wrap input in variable, wrap input in variable - no longer needed for
# v0.4 & higher code to handle PyTorch upgrade
pytorch_ver = __version__.split(".")
if int(pytorch_ver[0]) > 0 or int(pytorch_ver[1]) >= 4:
img_tensor.requires_grad_(False)
else:
data = Variable(img_tensor, volatile=True)
# apply model to input
model = models[model_name]
# puts model in evaluation mode
# instead of (default)training mode
model = model.eval()
if int(pytorch_ver[0])>0 or int(pytorch_ver[1])>=4:
output = model(img_tensor)
else:
output = model(data)
pred_idx = output.data.numpy().argmax()
return imagenet_classes_dict[pred_idx]
def classifier(img_path, model_name):
# load the image
img_pil = Image.open(img_path)
# define transforms
preprocess = transforms.Compose([
transforms.Resize(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
])
# preprocess the image
img_tensor = preprocess(img_pil)
img_tensor.unsqueeze_(0)
pytorch_ver = __version__.split('.')
if int(pytorch_ver[0]) > 0 or int(pytorch_ver[1]) >= 4:
img_tensor.requires_grad_(False)
# pytorch versions less than 0.4 - uses Variable because not-depreciated
else:
# apply model to input
# wrap input in variable
data = Variable(img_tensor, volatile=True)
# apply model to input
model = models[model_name]
model = model.eval()
# apply data to model - adjusted based upon version to account for
# operating on a Tensor for version 0.4 & higher.
if int(pytorch_ver[0]) > 0 or int(pytorch_ver[1]) >= 4:
output = model(img_tensor)
# pytorch versions less than 0.4
else:
# apply data to model
output = model(data)
pred_idx = output.data.numpy().argmax()
return imagenet_classes_dict[pred_idx]
config = ConfigParser()
rootdir = os.path.dirname(os.path.abspath(__file__))
config.read(os.path.join(rootdir, 'setup.cfg'))
INSTALL_REQUIRES = config['options']['install_requires']
INSTALL_REQUIRES = INSTALL_REQUIRES.split('\n')
COMPILED_BACKEND = os.environ.get('NI_COMPILED_BACKEND', '')
if not COMPILED_BACKEND:
warn(c_deprecation_warning, DeprecationWarning)
COMPILED_BACKEND = 'TS'
if COMPILED_BACKEND.upper() == 'C':
from setup_cext import prepare_extensions
from buildtools import build_ext
from torch import __version__ as torch_version
torch_version = torch_version.split('.')
if '.'.join(torch_version[:2]) == '1.7':
torch_version = '.'.join(torch_version[:3]) # we need the patch
else:
torch_version = '.'.join(torch_version[:3])
PYTORCH_TARGET = os.environ.get('NI_PYTORCH_TARGET', '')
SETUP_KWARGS['ext_package'] = 'nitorch'
SETUP_KWARGS['ext_modules'] = prepare_extensions()
CMDCLASS.update({'build_ext': build_ext})
INSTALL_REQUIRES += [f'torch=={torch_version}']
if COMPILED_BACKEND.upper() == 'MONAI':
INSTALL_REQUIRES += ['monai>=0.5']
SETUP_KWARGS['install_requires'] = INSTALL_REQUIRES
SETUP_KWARGS['cmdclass'] = CMDCLASS
#
# Copyright (C) Mellanox Technologies Ltd. 2001-2020. ALL RIGHTS RESERVED.
#
import os
import sys
from setuptools import setup
from torch.utils import cpp_extension
from torch import __version__ as torch_version
ver_major, ver_minor = torch_version.split(".")[:2]
ver_major = '-DTORCH_VER_MAJOR=' + ver_major
ver_minor = '-DTORCH_VER_MINOR=' + ver_minor
ucc_plugin_dir = os.path.dirname(os.path.abspath(__file__))
ucx_home = os.environ.get("UCX_HOME")
if ucx_home is None:
ucx_home = os.environ.get("HPCX_UCX_DIR")
if ucx_home is None:
print("Couldn't find UCX install dir, please set UCX_HOME env variable")
sys.exit(1)
plugin_sources = [
"src/torch_ucc.cpp", "src/torch_ucc_sendrecv.cpp",
"src/torch_ucx_alltoall.cpp", "src/torch_ucx_coll.cpp"
]
plugin_include_dirs = [
"{}/include/".format(ucc_plugin_dir), "{}/include/".format(ucx_home)
]
plugin_library_dirs = ["{}/lib/".format(ucx_home)]
plugin_libraries = ["ucp", "uct", "ucm", "ucs"]
