Big re-org, working towards making pip/module as 'timm'

inference.py

@@ -11,9 +11,9 @@ import argparse
 import numpy as np
 import torch
 
-from models import create_model, apply_test_time_pool
-from data import Dataset, create_loader, resolve_data_config
-from utils import AverageMeter
+from timm.models import create_model, apply_test_time_pool
+from timm.data import Dataset, create_loader, resolve_data_config
+from timm.utils import AverageMeter
 
 torch.backends.cudnn.benchmark = True
 
@@ -55,6 +55,9 @@ parser.add_argument('--topk', default=5, type=int,
 def main():
     args = parser.parse_args()
 
+    # might as well try to do something useful...
+    args.pretrained = args.pretrained or not args.checkpoint
+
     # create model
     model = create_model(
         args.model,

requirements.txt

@@ -0,0 +1,2 @@
+torch~=1.0
+torchvision

setup.py

@@ -0,0 +1,55 @@
+""" Setup
+"""
+from setuptools import setup, find_packages
+from codecs import open
+from os import path
+
+here = path.abspath(path.dirname(__file__))
+
+# Get the long description from the README file
+with open(path.join(here, 'README.md'), encoding='utf-8') as f:
+    long_description = f.read()
+
+exec(open('timm/version.py').read())
+setup(
+    name='timm',
+    version=__version__,
+    description='(Unofficial) PyTorch Image Models',
+    long_description=long_description,
+    url='https://github.com/rwightman/pytorch-image-models',
+    author='Ross Wightman',
+    author_email='hello@rwightman.com',
+    classifiers=[  # Optional
+        # How mature is this project? Common values are
+        #   3 - Alpha
+        #   4 - Beta
+        #   5 - Production/Stable
+        'Development Status :: 3 - Alpha',
+        'Intended Audience :: Developers',
+        'Topic :: Software Development :: Build Tools',
+        'License :: OSI Approved :: Apache License',
+        'Programming Language :: Python :: 3.6',
+    ],
+
+    # Note that this is a string of words separated by whitespace, not a list.
+    keywords='pytorch pretrained models efficientnet mobilenetv3 mnasnet',
+
+    # You can just specify package directories manually here if your project is
+    # simple. Or you can use find_packages().
+    #
+    # Alternatively, if you just want to distribute a single Python file, use
+    # the `py_modules` argument instead as follows, which will expect a file
+    # called `my_module.py` to exist:
+    #
+    #   py_modules=["my_module"],
+    #
+    packages=find_packages(exclude=['convert']),
+
+    # This field lists other packages that your project depends on to run.
+    # Any package you put here will be installed by pip when your project is
+    # installed, so they must be valid existing projects.
+    #
+    # For an analysis of "install_requires" vs pip's requirements files see:
+    # https://packaging.python.org/en/latest/requirements.html
+    install_requires=['torch', 'torchvision'],
+)

timm/__init__.py

@@ -0,0 +1,2 @@
+from .version import __version__
+from .models import create_model

timm/data/config.py

@@ -1,4 +1,4 @@
-from data.constants import *
+from .constants import *
 
 
 def resolve_data_config(model, args, default_cfg={}, verbose=True):

timm/data/loader.py

@@ -1,7 +1,7 @@
 import torch.utils.data
-from data.transforms import *
-from data.distributed_sampler import OrderedDistributedSampler
-from data.mixup import FastCollateMixup
+from .transforms import *
+from .distributed_sampler import OrderedDistributedSampler
+from .mixup import FastCollateMixup
 
 
 def fast_collate(batch):
@@ -101,7 +101,7 @@ def create_loader(
         img_size = input_size
 
     if tf_preprocessing and use_prefetcher:
-        from data.tf_preprocessing import TfPreprocessTransform
+        from timm.data.tf_preprocessing import TfPreprocessTransform
         transform = TfPreprocessTransform(is_training=is_training, size=img_size)
     else:
         if is_training:

timm/data/random_erasing.py

@@ -1,5 +1,3 @@
-from __future__ import absolute_import
-
 import random
 import math
 import torch

timm/data/transforms.py

@@ -7,8 +7,8 @@ import math
 import random
 import numpy as np
 
-from data import DEFAULT_CROP_PCT, IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD
-from data.random_erasing import RandomErasing
+from .constants import DEFAULT_CROP_PCT, IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD
+from .random_erasing import RandomErasing
 
 
 class ToNumpy:

timm/models/densenet.py

@@ -2,14 +2,11 @@
 This file is a copy of https://github.com/pytorch/vision 'densenet.py' (BSD-3-Clause) with
 fixed kwargs passthrough and addition of dynamic global avg/max pool.
 """
-import torch
-import torch.nn as nn
-import torch.nn.functional as F
 from collections import OrderedDict
 
-from models.helpers import load_pretrained
-from models.adaptive_avgmax_pool import *
-from data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD
+from .helpers import load_pretrained
+from .adaptive_avgmax_pool import *
+from timm.data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD
 import re
 
 _models = ['densenet121', 'densenet169', 'densenet201', 'densenet161']

timm/models/dpn.py

@@ -9,15 +9,13 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-import os
 import torch
 import torch.nn as nn
-import torch.nn.functional as F
 from collections import OrderedDict
 
-from models.helpers import load_pretrained
-from models.adaptive_avgmax_pool import select_adaptive_pool2d
-from data import IMAGENET_DPN_MEAN, IMAGENET_DPN_STD
+from .helpers import load_pretrained
+from .adaptive_avgmax_pool import select_adaptive_pool2d
+from timm.data import IMAGENET_DPN_MEAN, IMAGENET_DPN_STD
 
 _models = ['dpn68', 'dpn68b', 'dpn92', 'dpn98', 'dpn131', 'dpn107']
 __all__ = ['DPN'] + _models

timm/models/gen_efficientnet.py

@@ -22,10 +22,10 @@ from copy import deepcopy
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
-from models.helpers import load_pretrained
-from models.adaptive_avgmax_pool import SelectAdaptivePool2d
-from models.conv2d_same import sconv2d
-from data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD
+from .helpers import load_pretrained
+from .adaptive_avgmax_pool import SelectAdaptivePool2d
+from .conv2d_same import sconv2d
+from timm.data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD
 
 _models = [
     'mnasnet_050', 'mnasnet_075', 'mnasnet_100', 'mnasnet_b1', 'mnasnet_140', 'semnasnet_050', 'semnasnet_075',

timm/models/gluon_resnet.py

@@ -3,13 +3,12 @@ This file evolved from https://github.com/pytorch/vision 'resnet.py' with (SE)-R
 and ports of Gluon variations (https://github.com/dmlc/gluon-cv/blob/master/gluoncv/model_zoo/resnet.py) 
 by Ross Wightman
 """
-import torch
 import torch.nn as nn
 import torch.nn.functional as F
 import math
-from models.helpers import load_pretrained
-from models.adaptive_avgmax_pool import SelectAdaptivePool2d
-from data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD
+from .helpers import load_pretrained
+from .adaptive_avgmax_pool import SelectAdaptivePool2d
+from timm.data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD
 
 _models = [
     'gluon_resnet18_v1b', 'gluon_resnet34_v1b', 'gluon_resnet50_v1b', 'gluon_resnet101_v1b', 'gluon_resnet152_v1b',

timm/models/inception_resnet_v2.py

@@ -2,12 +2,9 @@
 Sourced from https://github.com/Cadene/tensorflow-model-zoo.torch (MIT License) which is
 based upon Google's Tensorflow implementation and pretrained weights (Apache 2.0 License)
 """
-import torch
-import torch.nn as nn
-import torch.nn.functional as F
-from models.helpers import load_pretrained
-from models.adaptive_avgmax_pool import *
-from data import IMAGENET_INCEPTION_MEAN, IMAGENET_INCEPTION_STD
+from .helpers import load_pretrained
+from .adaptive_avgmax_pool import *
+from timm.data import IMAGENET_INCEPTION_MEAN, IMAGENET_INCEPTION_STD
 
 _models = ['inception_resnet_v2']
 __all__ = ['InceptionResnetV2'] + _models

timm/models/inception_v3.py

@@ -1,6 +1,6 @@
 from torchvision.models import Inception3
-from models.helpers import load_pretrained
-from data import IMAGENET_DEFAULT_STD, IMAGENET_DEFAULT_MEAN, IMAGENET_INCEPTION_MEAN, IMAGENET_INCEPTION_STD
+from .helpers import load_pretrained
+from timm.data import IMAGENET_DEFAULT_STD, IMAGENET_DEFAULT_MEAN, IMAGENET_INCEPTION_MEAN, IMAGENET_INCEPTION_STD
 
 _models = ['inception_v3', 'tf_inception_v3', 'adv_inception_v3', 'gluon_inception_v3']
 __all__ = _models

timm/models/inception_v4.py

@@ -2,12 +2,9 @@
 Sourced from https://github.com/Cadene/tensorflow-model-zoo.torch (MIT License) which is
 based upon Google's Tensorflow implementation and pretrained weights (Apache 2.0 License)
 """
-import torch
-import torch.nn as nn
-import torch.nn.functional as F
-from models.helpers import load_pretrained
-from models.adaptive_avgmax_pool import *
-from data import IMAGENET_INCEPTION_MEAN, IMAGENET_INCEPTION_STD
+from .helpers import load_pretrained
+from .adaptive_avgmax_pool import *
+from timm.data import IMAGENET_INCEPTION_MEAN, IMAGENET_INCEPTION_STD
 
 _models = ['inception_v4']
 __all__ = ['InceptionV4'] + _models

timm/models/model_factory.py

@@ -1,21 +1,21 @@
-from models.inception_v4 import *
-from models.inception_resnet_v2 import *
-from models.densenet import *
-from models.resnet import *
-from models.dpn import *
-from models.senet import *
-from models.xception import *
-from models.pnasnet import *
-from models.gen_efficientnet import *
-from models.inception_v3 import *
-from models.gluon_resnet import *
+from .inception_v4 import *
+from .inception_resnet_v2 import *
+from .densenet import *
+from .resnet import *
+from .dpn import *
+from .senet import *
+from .xception import *
+from .pnasnet import *
+from .gen_efficientnet import *
+from .inception_v3 import *
+from .gluon_resnet import *
 
-from models.helpers import load_checkpoint
+from .helpers import load_checkpoint
 
 
 def create_model(
-        model_name='resnet50',
-        pretrained=None,
+        model_name,
+        pretrained=False,
         num_classes=1000,
         in_chans=3,
         checkpoint_path='',

timm/models/pnasnet.py

@@ -12,8 +12,8 @@ import torch
 import torch.nn as nn
 import torch.nn.functional as F
 
-from models.helpers import load_pretrained
-from models.adaptive_avgmax_pool import SelectAdaptivePool2d
+from .helpers import load_pretrained
+from .adaptive_avgmax_pool import SelectAdaptivePool2d
 
 _models = ['pnasnet5large']
 __all__ = ['PNASNet5Large'] + _models

timm/models/resnet.py

@@ -4,13 +4,12 @@ additional dropout and dynamic global avg/max pool.
 
 ResNext additions added by Ross Wightman
 """
-import torch
 import torch.nn as nn
 import torch.nn.functional as F
 import math
-from models.helpers import load_pretrained
-from models.adaptive_avgmax_pool import SelectAdaptivePool2d
-from data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD
+from .helpers import load_pretrained
+from .adaptive_avgmax_pool import SelectAdaptivePool2d
+from timm.data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD
 
 _models = ['resnet18', 'resnet34', 'resnet50', 'resnet101', 'resnet152',
            'resnext50_32x4d', 'resnext101_32x4d', 'resnext101_64x4d', 'resnext152_32x4d']

timm/models/senet.py

@@ -15,9 +15,9 @@ import math
 import torch.nn as nn
 import torch.nn.functional as F
 
-from models.helpers import load_pretrained
-from models.adaptive_avgmax_pool import SelectAdaptivePool2d
-from data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD
+from .helpers import load_pretrained
+from .adaptive_avgmax_pool import SelectAdaptivePool2d
+from timm.data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD
 
 _models = ['seresnet18', 'seresnet34', 'seresnet50', 'seresnet101', 'seresnet152', 'senet154',
            'seresnext26_32x4d', 'seresnext50_32x4d', 'seresnext101_32x4d']

timm/models/test_time_pool.py

@@ -1,6 +1,6 @@
 from torch import nn
 import torch.nn.functional as F
-from models.adaptive_avgmax_pool import adaptive_avgmax_pool2d
+from .adaptive_avgmax_pool import adaptive_avgmax_pool2d
 
 
 class TestTimePoolHead(nn.Module):

timm/models/xception.py

@@ -27,8 +27,8 @@ import torch
 import torch.nn as nn
 import torch.nn.functional as F
 
-from models.helpers import load_pretrained
-from models.adaptive_avgmax_pool import select_adaptive_pool2d
+from .helpers import load_pretrained
+from .adaptive_avgmax_pool import select_adaptive_pool2d
 
 _models = ['xception']
 __all__ = ['Xception'] + _models

timm/optim/optim_factory.py

@@ -1,5 +1,5 @@
 from torch import optim as optim
-from optim import Nadam, RMSpropTF
+from timm.optim import Nadam, RMSpropTF
 
 
 def add_weight_decay(model, weight_decay=1e-5, skip_list=()):

timm/scheduler/scheduler_factory.py

@@ -1,7 +1,6 @@
-from scheduler.cosine_lr import CosineLRScheduler
-from scheduler.plateau_lr import PlateauLRScheduler
-from scheduler.tanh_lr import TanhLRScheduler
-from scheduler.step_lr import StepLRScheduler
+from .cosine_lr import CosineLRScheduler
+from .tanh_lr import TanhLRScheduler
+from .step_lr import StepLRScheduler
 
 
 def create_scheduler(args, optimizer):

timm/version.py

@@ -0,0 +1 @@
+__version__ = '0.1.1'

train.py

@@ -11,16 +11,15 @@ try:
 except ImportError:
     has_apex = False
 
-from data import Dataset, create_loader, resolve_data_config, FastCollateMixup, mixup_target
-from models import create_model, resume_checkpoint, load_checkpoint
-from utils import *
-from loss import LabelSmoothingCrossEntropy, SoftTargetCrossEntropy
-from optim import create_optimizer
-from scheduler import create_scheduler
+from timm.data import Dataset, create_loader, resolve_data_config, FastCollateMixup, mixup_target
+from timm.models import create_model, resume_checkpoint
+from timm.utils import *
+from timm.loss import LabelSmoothingCrossEntropy, SoftTargetCrossEntropy
+from timm.optim import create_optimizer
+from timm.scheduler import create_scheduler
 
 import torch
 import torch.nn as nn
-import torch.distributed as dist
 import torchvision.utils
 
 torch.backends.cudnn.benchmark = True

validate.py

@@ -12,9 +12,9 @@ import torch.nn as nn
 import torch.nn.parallel
 from collections import OrderedDict
 
-from models import create_model, apply_test_time_pool, load_checkpoint
-from data import Dataset, create_loader, resolve_data_config
-from utils import accuracy, AverageMeter, natural_key
+from timm.models import create_model, apply_test_time_pool, load_checkpoint
+from timm.data import Dataset, create_loader, resolve_data_config
+from timm.utils import accuracy, AverageMeter, natural_key
 
 torch.backends.cudnn.benchmark = True