class CLIConfig(scfg.Config):
"""
Create COCO toydata
"""
default = {
'key':
scfg.Value('shapes8', help='special demodata code', position=1),
'dst':
scfg.Value(None,
help=ub.paragraph('''
Output path for the final kwcoco json file.
Note, that even when given, a data.kwcoco.json file
will also be generated in a bundle_dpath.
''')),
'bundle_dpath':
scfg.Value(None,
help=ub.paragraph('''
Creates a bundled dataset in the specified location.
If unspecified, a bundle name is generated based on the
toydata config.
''')),
'use_cache':
scfg.Value(True)
}
epilog = """
class ExampleConfig(scfg.Config):
default = {
'item1': [],
'item2': scfg.Value([], type=list),
'item3': scfg.Value([]),
'item4': scfg.Value([], nargs='*'),
}
class ListDeployedConfig(scfg.Config):
"""
Given a netharn work directory list all deployed models
"""
default = {
'workdir': scfg.Value(None, help='work directory'),
'name': scfg.Value(None, help='"nice" name of the run'),
}
class ConvertConfig(scfg.Config):
default = {
'src': scfg.PathList('in.viame.csv'),
'dst': scfg.Value('out.kwcoco.json'),
'new_root': None,
'old_root': None,
'images': scfg.Value(None, help='image list file or path to image directory if the CSV does not specify image names'),
}
class MyConfig(scfg.Config):
default = {
'option1': scfg.Value('bar', help='an option'),
'option2': scfg.Value((1, 2, 3), tuple, help='another option'),
'option3': None,
'option4': 'foo',
'discrete': scfg.Value(None, choices=['a', 'b', 'c']),
'apath': scfg.Path(help='a path'),
}
class CLIConfig(scfg.Config):
"""
Combine multiple COCO datasets into a single merged dataset.
"""
default = {
'src': scfg.Value([], nargs='+', help='path to multiple input datasets', position=1),
'dst': scfg.Value('combo.mscoco.json', help='path to output dataset'),
'absolute': scfg.Value(False, help='if True, converts paths to absolute paths before doing union')
}
epilog = """
class DetectPredictCLIConfig(scfg.Config):
default = ub.dict_union(
{
'dataset': scfg.Value(None, help='coco dataset, path to images or folder of images'),
'out_dpath': scfg.Value('./out', help='output directory'),
'draw': scfg.Value(False),
'workdir': scfg.Value('~/work/bioharn', help='work directory for sampler if needed'),
},
DetectPredictConfig.default
)
class CLIConfig(scfg.Config):
"""
Compute summary statistics about a COCO dataset
"""
default = {
'src':
scfg.Value(['special:shapes8'],
nargs='+',
help='path to dataset',
position=1),
'basic':
scfg.Value(True, help='show basic stats'),
'extended':
scfg.Value(True, help='show extended stats'),
'catfreq':
scfg.Value(True, help='show category frequency stats'),
'boxes':
scfg.Value(False,
help=ub.paragraph('''
show bounding box stats in width-height format.
''')),
'annot_attrs':
scfg.Value(False, help='show annotation attribute information'),
'image_attrs':
scfg.Value(False, help='show image attribute information'),
'video_attrs':
scfg.Value(False, help='show video attribute information'),
'embed':
scfg.Value(False, help='embed into interactive shell'),
}
epilog = """
class CLIConfig(scfg.Config):
"""
Split a single COCO dataset into two sub-datasets.
"""
default = {
'src': scfg.Value(None, help='input dataset to split', position=1),
'dst1': scfg.Value('split1.mscoco.json', help='output path1'),
'dst2': scfg.Value('split2.mscoco.json', help='output path2'),
'factor': scfg.Value(3,
help='ratio of items put in dset1 vs dset2'),
'rng': scfg.Value(None, help='random seed'),
}
epilog = """
class CLIConfig(scfg.Config):
"""
Create a COCO file from bitmasks
"""
default = {
'src':
scfg.PathList(
help='a file, globstr, or comma-separated list of files'),
'dst':
scfg.Value('masks.mscoco.json', help='output path'),
'serialization':
scfg.Value('vector', help='can be raster or vector'),
}
epilog = r"""
class FileHashConfig(scfg.Config):
"""
The docstring will be the description in the CLI help
"""
default = {
'fpath': scfg.Value(None, position=1, help=ub.paragraph(
'''
a path to a file to hash
''')),
'hasher': scfg.Value('sha1', choices=['sha1', 'sha512'], help=ub.paragraph(
'''
a name of a hashlib hasher'
''')),
}
class DemoConfig(scfg.Config):
default = {
'num': 1,
'mode': 'bar',
'mode2': scfg.Value('bar', str),
'ignore': ['baz', 'biz'],
}
class DocstrStubgenCLI(scfg.Config):
name = 'doctypes'
description = 'Generate Typed Stubs from Docstrings'
default = {
'module':
scfg.Value(None,
position=1,
help=ub.paragraph('''
The name of a module in the PYTHONPATH or an explicit path to that
module.
''')),
}
@classmethod
def main(cls, cmdline=False, **kwargs):
from xdev.cli import docstr_stubgen
import ubelt as ub
config = cls(cmdline=cmdline, data=kwargs)
print(f'config={config}')
modname_or_path = config['module']
print(f'modname_or_path={modname_or_path}')
if modname_or_path is None:
raise ValueError('Must specify the module')
modpath = docstr_stubgen.modpath_coerce(modname_or_path)
modpath = ub.Path(modpath)
generated = docstr_stubgen.generate_typed_stubs(modpath)
for fpath, text in generated.items():
fpath = ub.Path(fpath)
print(f'Write fpath={fpath}')
fpath.write_text(text)
# Generate a py.typed file to mark the package as typed
if modpath.is_dir():
pytyped_fpath = (modpath / 'py.typed')
print(f'touch pytyped_fpath={pytyped_fpath}')
pytyped_fpath.touch()
class SedCLI(scfg.Config):
name = 'sed'
description = 'Search and replace text in files'
default = {
'regexpr':
scfg.Value('',
position=1,
help=ub.paragraph('''
The pattern to search for.
''')),
'repl':
scfg.Value('',
position=2,
help=ub.paragraph('''
The pattern to replace with.
''')),
'dpath':
scfg.Value(None,
position=3,
help=ub.paragraph('''
The directory to recursively search or a file pattern to match.
''')),
'dry':
scfg.Value('ask',
position=4,
help=ub.paragraph('''
if 1, show what would be done. if 0, execute the change, if "ask",
then show the dry run and then ask for confirmation.
''')),
'include':
scfg.Value(None),
'exclude':
scfg.Value(None),
'recursive':
scfg.Value(True),
'verbose':
scfg.Value(1),
}
@classmethod
def main(cls, cmdline=False, **kwargs):
from xdev import search_replace
config = cls(cmdline=cmdline, data=kwargs)
if config['dry'] in {'ask', 'auto'}:
from rich.prompt import Confirm
config['dry'] = True
search_replace.sed(**config)
flag = Confirm.ask('Do you want to execute this sed?')
if flag:
config['dry'] = False
search_replace.sed(**config)
else:
search_replace.sed(**config)
class CodeblockCLI(scfg.Config):
name = 'codeblock'
description = 'Remove indentation from text'
default = {
'text': scfg.Value('', position=1, help='text to dedent'),
}
@classmethod
def main(cls, cmdline=False, **kwargs):
config = cls(cmdline=cmdline, data=kwargs)
print(ub.codeblock(config['text']))
class DetectPredictConfig(scfg.Config):
default = {
'deployed': None,
'batch_size': 4,
'xpu': 'auto',
'window_dims': scfg.Value('full', help='size of a sliding window'), # (512, 512),
'input_dims': scfg.Value((512, 512), help='The size of the inputs to the network'),
'workers': 0,
'overlap': scfg.Value(0.0, help='overlap of the sliding window'),
# Note: these dont work exactly correct due to mmdetection model
# differences
'nms_thresh': 0.4,
'conf_thresh': 0.1,
'verbose': 3,
}
class CLIConfig(scfg.Config):
"""
Reroot image paths onto a new image root.
"""
epilog = """
Example Usage:
kwcoco reroot --help
kwcoco reroot --src=special:shapes8 --dst rerooted.json
kwcoco reroot --src=special:shapes8 --new_prefix=foo --check=True --dst rerooted.json
"""
default = {
'src': scfg.Value(None, help=(
'Path to the coco dataset'), position=1),
'new_prefix': scfg.Value(None, help=(
'Path to the new image root.')),
'old_prefix': scfg.Value(None, help=(
'Previous root to remove.')),
'absolute': scfg.Value(True, help=(
'If False, the output file uses relative paths')),
'check': scfg.Value(True, help=(
'If True, checks that all data exists')),
'dst': scfg.Value(None, help=(
'Save the re-rooted dataset to a new file')),
}
class CLIConfig(scfg.Config):
"""
Make the COCO file conform to the spec.
Populates inferable information such as image size, annotation area, etc.
"""
epilog = """
Example Usage:
kwcoco conform --help
kwcoco conform --src=special:shapes8 --dst conformed.json
"""
default = {
'src':
scfg.Value(None, help=('Path to the coco dataset'), position=1),
'ensure_imgsize':
scfg.Value(True,
help=ub.paragraph('''
ensure each image has height and width attributes
''')),
'pycocotools_info':
scfg.Value(True,
help=ub.paragraph('''
ensure information needed for pycocotools
''')),
'legacy':
scfg.Value(False,
help='if True tries to convert to the '
'original ms-coco format'),
'workers':
scfg.Value(8,
help='number of background workers for bigger checks'),
'dst':
scfg.Value(None, help=('Save the modified dataset to a new file')),
}
class FindCLI(scfg.Config):
name = 'find'
description = 'Find files based on names'
default = {
'pattern': scfg.Value('', position=1),
'dpath': scfg.Value(None, position=2),
'include': scfg.Value(None),
'exclude': scfg.Value(None),
'type': scfg.Value('f', help="can be f and/or d"),
'recursive': scfg.Value(True),
'followlinks': scfg.Value(False),
}
@classmethod
def main(cls, cmdline=False, **kwargs):
from xdev import search_replace
config = cls(cmdline=cmdline, data=kwargs)
for found in search_replace.find(**config):
print(found)
class CLIConfig(scfg.Config):
"""
Visualize a COCO image using matplotlib or opencv, optionally writing
it to disk
"""
epilog = """
Example Usage:
kwcoco show --help
kwcoco show --src=special:shapes8 --gid=1
kwcoco show --src=special:shapes8 --gid=1 --dst out.png
"""
default = {
'src':
scfg.Value(None, help=('Path to the coco dataset'), position=1),
'gid':
scfg.Value(
None,
help=(
'Image id to show, if unspecified the first image is shown'
)),
'aid':
scfg.Value(
None,
help=('Annotation id to show, mutually exclusive with gid')),
'dst':
scfg.Value(
None,
help=('Save the image to the specified file. '
'If unspecified, the image is shown with pyplot')),
'mode':
scfg.Value('matplotlib',
choices=['matplotlib', 'opencv'],
help='method used to draw the image'),
'channels':
scfg.Value(None,
type=str,
help=ub.paragraph('''
By default uses the default channels (usually this is rgb),
otherwise specify the name of an auxiliary channels
''')),
'show_annots':
scfg.Value(True, help=('Overlay annotations on dispaly')),
}
class CLIConfig(scfg.Config):
"""
Rename or remove categories
"""
epilog = """
Example Usage:
kwcoco modify_categories --help
kwcoco modify_categories --src=special:shapes8 --dst modcats.json
kwcoco modify_categories --src=special:shapes8 --dst modcats.json --rename eff:F,star:sun
kwcoco modify_categories --src=special:shapes8 --dst modcats.json --remove eff,star
kwcoco modify_categories --src=special:shapes8 --dst modcats.json --keep eff,
kwcoco modify_categories --src=special:shapes8 --dst modcats.json --keep=[] --keep_annots=True
"""
default = {
'src':
scfg.Value(None, help=('Path to the coco dataset'), position=1),
'dst':
scfg.Value(None, help=('Save the rebased dataset to a new file')),
'keep_annots':
scfg.Value(
False,
help=
('if False, removes annotations when categories are removed, '
'otherwise the annotations category is simply unset')),
'remove':
scfg.Value(None,
help='Category names to remove. Mutex with keep.'),
'keep':
scfg.Value(
None,
help=
'If specified, remove all other categories. Mutex with remove.'
),
'rename':
scfg.Value(
None,
type=str,
help='category mapping in the format. "old1:new1,old2:new2"'),
}
class CLIConfig(scfg.Config):
"""
Validate that a coco file conforms to the json schema, that assets
exist, and potentially fix corrupted assets by removing them.
"""
default = {
'src':
scfg.Value(['special:shapes8'],
nargs='+',
help='path to datasets',
position=1),
'schema':
scfg.Value(True, help='If True check the json schema'),
'missing':
scfg.Value(True,
help='If True check if all assets (e.g. images) exist'),
'corrupted':
scfg.Value(False, help='If True check the assets can be read'),
'fix':
scfg.Value(None,
help=ub.paragraph('''
Code indicating strategy to attempt to fix the dataset.
If None, do nothing.
If remove, removes missing / corrupted images.
Other strategies may be added in the future.
This is a hueristic and does not always work. dst must be
specified. And only one src dataset can be given.
''')),
'dst':
scfg.Value(None,
help=ub.paragraph('''
Location to write a "fixed" coco file if a fix strategy is
given.
'''))
}
epilog = """
class TestConfig(scfg.Config):
default = {
'key': scfg.Value(None, type=str),
}
class TemplateConfig(scfg.Config):
default = {
'repodir':
scfg.Value(None,
help='path to the new or existing repo',
required=True),
'repo_name':
scfg.Value(None, help='repo name'),
'setup_secrets':
scfg.Value(False),
'tags':
scfg.Value([],
nargs='*',
help=ub.paragraph('''
Tags modify what parts of the template are used.
Valid tags are:
"binpy" - do we build binpy wheels?
"graphics" - do we need opencv / opencv-headless?
"erotemic" - this is an erotemic repo
"kitware" - this is an kitware repo
"pyutils" - this is an pyutils repo
"purepy" - this is a pure python repo
''')),
}
def normalize(self):
if self['tags']:
if isinstance(self['tags'], str):
self['tags'] = [self['tags']]
new = []
for t in self['tags']:
new.extend([p.strip() for p in t.split(',')])
self['tags'] = new
@classmethod
def main(cls, cmdline=0, **kwargs):
"""
Ignore:
repodir = ub.Path('~/code/pyflann_ibeis').expand()
kwargs = {
'repodir': repodir,
'tags': ['binpy', 'erotemic', 'github'],
}
cmdline = 0
Example:
repodir = ub.Path.appdir('pypkg/demo/my_new_repo')
import sys, ubelt
sys.path.append(ubelt.expandpath('~/misc/templates/PYPKG'))
from apply_template import * # NOQA
kwargs = {
'repodir': repodir,
}
cmdline = 0
"""
import ubelt as ub
config = TemplateConfig(cmdline=cmdline, data=kwargs)
repo_dpath = ub.Path(config['repodir'])
repo_dpath.ensuredir()
IS_NEW_REPO = 0
create_new_repo_info = ub.codeblock('''
# TODO:
# At least instructions on how to create a new repo, or maybe an
# API call
https://github.com/new
git init
''')
print(create_new_repo_info)
if IS_NEW_REPO:
# TODO: git init
# TODO: github or gitlab register
pass
self = TemplateApplier(config)
self.setup().gather_tasks()
self.setup().apply()
if config['setup_secrets']:
setup_secrets_fpath = self.repo_dpath / 'dev/setup_secrets.sh'
if 'erotemic' in self.config['tags']:
environ_export = 'setup_package_environs_github_erotemic'
upload_secret_cmd = 'upload_github_secrets'
elif 'pyutils' in self.config['tags']:
environ_export = 'setup_package_environs_github_pyutils'
upload_secret_cmd = 'upload_github_secrets'
elif 'kitware' in self.config['tags']:
environ_export = 'setup_package_environs_gitlab_kitware'
upload_secret_cmd = 'upload_gitlab_repo_secrets'
else:
raise Exception
import cmd_queue
script = cmd_queue.Queue.create()
script.submit(
ub.codeblock(f'''
cd {self.repo_dpath}
source {setup_secrets_fpath}
{environ_export}
load_secrets
export_encrypted_code_signing_keys
git commit -am "Updated secrets"
{upload_secret_cmd}
'''))
script.rprint()
class ConddaConfig(scfg.Config):
"""
Default configuration for CONDDA protocol.
Example:
>>> from sail_on_client.protocol.condda_config import ConddaConfig
>>> config = ConddaConfig()
>>> print('config = {!r}'.format(config))
"""
default = {
"domain":
scfg.Value("image_classification"),
"test_ids": ["CONDDA.2.1.293"],
"novelty_detector_class":
scfg.Value("CONDDA_5_14_A1"),
"seed":
scfg.Value("seed"),
"dataset_root":
"",
"feature_extraction_only":
scfg.Value(False, help="Quit after feature extraction"),
"save_features":
scfg.Value(False, help="Save features as pkl file"),
"use_saved_features":
scfg.Value(False, help="Use features saved the pkl file"),
"save_dir":
scfg.Value("", help="Directory where features are saved"),
"save_attributes":
scfg.Value(False, help="Flag to attributes in save dir"),
"use_saved_attributes":
scfg.Value(False, help="Use attributes saved in save dir"),
"saved_attributes": {},
"skip_stage": [],
"hints": [],
"detector_config": {
"efficientnet_params": {
"model_path": "",
"known_classes": 413
},
"evm_params": {
"model_path": "",
"known_feature_path": "",
"tailsize": 33998,
"cover_threshold": 0.7,
"distance_multiplier": 0.55,
"number_of_unknown_to_crate_evm": 3,
},
"dataloader_params": {
"batch_size": 128,
"num_workers": 3
},
"characterization_param": {
"clustering_type": "FINCH",
"number_of_unknown_to_strat_clustering": 20,
},
"csv_folder": "",
"cores": 4,
"detection_threshold": 0.5,
},
}
class DetectFitConfig(scfg.Config):
default = {
# Personal Preference
'nice':
scfg.Value(
'untitled',
help=
('a human readable tag for your experiment (we also keep a '
'failsafe computer readable tag in case you update hyperparams, '
'but forget to update this flag)')),
# System Options
'workdir':
scfg.Path('~/work/detect',
help='path where this script can dump stuff'),
'workers':
scfg.Value(0, help='number of DataLoader processes'),
'xpu':
scfg.Value('argv', help='a CUDA device or a CPU'),
# Data (the hardest part of machine learning)
'datasets':
scfg.Value('special:shapes1024', help='special dataset key'),
'train_dataset':
scfg.Value(None, help='override train with a custom coco dataset'),
'vali_dataset':
scfg.Value(None, help='override vali with a custom coco dataset'),
'test_dataset':
scfg.Value(None, help='override test with a custom coco dataset'),
# Dataset options
'multiscale':
False,
'visible_thresh':
scfg.Value(
0.5,
help=
'percentage of a box that must be visible to be included in truth'
),
'input_dims':
scfg.Value((256, 256), help='size to '),
'normalize_inputs':
scfg.Value(
False,
help='if True, precompute training mean and std for data whitening'
),
'augment':
scfg.Value('simple',
help='key indicating augmentation strategy',
choices=['complex', 'simple', None]),
'ovthresh':
0.5,
# High level options
'arch':
scfg.Value('yolo2', help='network toplogy', choices=['yolo2']),
'optim':
scfg.Value('adam',
help='torch optimizer',
choices=['sgd', 'adam', 'adamw']),
'batch_size':
scfg.Value(
4, help='number of images that run through the network at a time'),
'bstep':
scfg.Value(8, help='num batches before stepping'),
'lr':
scfg.Value(1e-3, help='learning rate'), # 1e-4,
'decay':
scfg.Value(1e-5, help='weight decay'),
'schedule':
scfg.Value('step90', help='learning rate / momentum scheduler'),
'max_epoch':
scfg.Value(140, help='Maximum number of epochs'),
'patience':
scfg.Value(
140,
help='Maximum number of bad epochs on validation before stopping'),
# Initialization
'init':
scfg.Value('imagenet', help='initialization strategy'),
'pretrained':
scfg.Path(help='path to a netharn deploy file'),
# Loss Terms
'focus':
scfg.Value(0.0, help='focus for Focal Loss'),
}
def normalize(self):
if self['pretrained'] in ['null', 'None']:
self['pretrained'] = None
if self['datasets'] == 'special:voc':
self['train_dataset'] = ub.expandpath(
'~/data/VOC/voc-trainval.mscoco.json')
self['vali_dataset'] = ub.expandpath(
'~/data/VOC/voc-test-2007.mscoco.json')
key = self.get('pretrained', None) or self.get('init', None)
if key == 'imagenet':
self['pretrained'] = yolo2.initial_imagenet_weights()
elif key == 'lightnet':
self['pretrained'] = yolo2.demo_voc_weights()
if self['pretrained'] is not None:
self['init'] = 'pretrained'
class TestConfig(scfg.Config):
default = {
'paths': scfg.Value(None, nargs='+'),
}
class ConddaConfig(scfg.Config):
"""
Default configuration for CONDDA protocol.
Example:
>>> from sail_on_client.protocol.condda_config import ConddaConfig
>>> config = ConddaConfig()
>>> print('config = {!r}'.format(config))
"""
default = {
"domain":
scfg.Value("image_classification"),
"test_ids": ["CONDDA.2.1.293"],
"seed":
scfg.Value("seed"),
"dataset_root":
"",
"feature_extraction_only":
scfg.Value(False, help="Quit after feature extraction"),
"save_features":
scfg.Value(False, help="Save features as pkl file"),
"use_saved_features":
scfg.Value(False, help="Use features saved the pkl file"),
"use_consolidated_features":
scfg.Value(False,
help="Use features consolidated over multiple tests"),
"save_dir":
scfg.Value("", help="Directory where features are saved"),
"save_attributes":
scfg.Value(False, help="Flag to attributes in save dir"),
"use_saved_attributes":
scfg.Value(False, help="Use attributes saved in save dir"),
"saved_attributes": {},
"skip_stage": [],
"hints": [],
"resume_session":
scfg.Value(False, help="Flag to resume session"),
"resumed_session_ids": {},
"detectors": {
"has_baseline": False,
"has_reaction_baseline": False,
"baseline_class": None,
"csv_folder": "",
"cores": 6,
"detection_threshold": 0.1,
"detector_configs": {
"CONDDA_5_14_A1": {
"feature_extractor_params": {
"backbone_weight_path": "",
"name": "i3d",
"arch": "i3d-50",
"graph_weight_path": "",
"model_name": "i3d",
"n_classes": 400,
"no_cuda": "False",
"hidden_dims": [512, 128],
"hidden": "True",
"in_dim": 1024,
"num_heads": [4, 1],
"sample_duration": 64,
"graph_classes": 88,
"mode": "feature",
"feature_type": "graph",
},
"evm_params": {
"weight_path": "",
"number_of_unknown_to_crate_evm": 7,
},
"characterization_params": {
"clustering_type": "FINCH",
"number_of_unknown_to_strat_clustering": 50,
},
"dataloader_params": {
"sample_size": 224,
"mean": [114.7748, 107.7354, 99.4750],
"sample_duration": 64,
"batch_size": 1,
"n_threads": 6,
"n_classes": 88,
},
},
},
},
"harness_config": {
"url": "http://3.32.8.161:5001/",
"data_location": "",
"data_dir": "",
"gt_dir": "",
"gt_config": "",
},
}
class OndConfig(scfg.Config):
"""
Default configuration for Ond protocol.
Example:
>>> from learn.protocol.learn_config import LearnConfig
>>> config = LearnConfig()
>>> print('config = {!r}'.format(config))
"""
default = {
"domain":
scfg.Value("image_classification"),
"test_ids": ["OND.1.1.293"],
"seed":
scfg.Value("seed"),
"dataset_root":
"/home/eric/sail-on/images",
"feature_extraction_only":
scfg.Value(False, help="Quit after feature extraction"),
"use_feedback":
scfg.Value(False, help="Use feedback for the run"),
"feedback_type":
scfg.Value("classification", help="Type of feedback"),
"save_features":
scfg.Value(False, help="Save features as pkl file"),
"use_saved_features":
scfg.Value(False, help="Use features saved the pkl file"),
"use_consolidated_features":
scfg.Value(False,
help="Use features consolidated over multiple tests"),
"save_dir":
scfg.Value("", help="Directory where features are saved"),
"save_attributes":
scfg.Value(False, help="Flag to attributes in save dir"),
"use_saved_attributes":
scfg.Value(False, help="Use attributes saved in save dir"),
"save_elementwise":
scfg.Value(False, help="Save attributes elementwise"),
"is_eval_enabled":
scfg.Value(False, help="Flag to enable evaluate"),
"is_eval_roundwise_enabled":
scfg.Value(False, help="Flag to enable roundwise evaluate"),
"saved_attributes": {},
"skip_stage": [],
"hints": [],
"resume_session":
scfg.Value(False, help="Flag to resume session"),
"resumed_session_ids": {},
"detectors": {
"has_baseline": False,
"has_reaction_baseline": False,
"baseline_class": None,
"csv_folder": "",
"cores": 6,
"detection_threshold": 0.1,
"detector_configs": {
"gae_kl_nd": {
"feature_extractor_params": {
"backbone_weight_path": "",
"name": "i3d",
"arch": "i3d-50",
"graph_weight_path": "",
"model_name": "i3d",
"n_classes": 400,
"no_cuda": "False",
"hidden_dims": [512, 128],
"hidden": "True",
"in_dim": 1024,
"num_heads": [4, 1],
"sample_duration": 64,
"graph_classes": 88,
"mode": "feature",
"feature_type": "graph",
},
"kl_params": {
"window_size": 100,
"mu_train": 1.0,
"sigma_train": 0.0057400320777888664,
"KL_threshold": 5.830880886275709,
},
"evm_params": {
"weight_path": "",
"number_of_unknown_to_crate_evm": 7,
},
"characterization_params": {
"clustering_type": "FINCH",
"number_of_unknown_to_strat_clustering": 50,
},
"dataloader_params": {
"sample_size": 224,
"mean": [114.7748, 107.7354, 99.4750],
"sample_duration": 64,
"batch_size": 1,
"n_threads": 6,
"n_classes": 88,
},
},
"baseline_i3d": {
"feature_extractor_params": {
"backbone_weight_path": "",
"name": "i3d",
"arch": "i3d-50",
"graph_weight_path": "",
"model_name": "i3d",
"n_classes": 400,
"no_cuda": "False",
"hidden_dims": [512, 128],
"hidden": "True",
"in_dim": 1024,
"num_heads": [4, 1],
"sample_duration": 64,
"graph_classes": 88,
"mode": "feature",
"feature_type": "graph",
},
},
},
},
"harness_config": {
"url": "http://3.32.8.161:5001/",
"data_location": "",
"data_dir": "",
"gt_dir": "",
"gt_config": "",
},
}
class SegmentationConfig(scfg.Config):
"""
Default configuration for setting up a training session
"""
default = {
'nice': scfg.Path('untitled', help='A human readable tag that is "nice" for humans'),
'workdir': scfg.Path('~/work/camvid', help='Dump all results in your workdir'),
'workers': scfg.Value(0, help='number of parallel dataloading jobs'),
'xpu': scfg.Value('argv', help='See netharn.XPU for details. can be cpu/gpu/cuda0/0,1,2,3)'),
'augment': scfg.Value('simple', help='type of training dataset augmentation'),
'class_weights': scfg.Value('log-median-idf', help='how to weight inbalanced classes'),
# 'class_weights': scfg.Value(None, help='how to weight inbalanced classes'),
'datasets': scfg.Value('special:camvid', help='Eventually you may be able to sepcify a coco file'),
'train_dataset': scfg.Value(None),
'vali_dataset': scfg.Value(None),
'arch': scfg.Value('psp', help='Network architecture code'),
'optim': scfg.Value('adamw', help='Weight optimizer. Can be SGD, ADAM, ADAMW, etc..'),
'input_dims': scfg.Value((128, 128), help='Window size to input to the network'),
'input_overlap': scfg.Value(0.25, help='amount of overlap when creating a sliding window dataset'),
'batch_size': scfg.Value(4, help='number of items per batch'),
'bstep': scfg.Value(1, help='number of batches before a gradient descent step'),
'max_epoch': scfg.Value(140, help='Maximum number of epochs'),
'patience': scfg.Value(140, help='Maximum "bad" validation epochs before early stopping'),
'lr': scfg.Value(1e-3, help='Base learning rate'),
'decay': scfg.Value(1e-5, help='Base weight decay'),
'focus': scfg.Value(2.0, help='focus for focal loss'),
'schedule': scfg.Value('step90', help=('Special coercable netharn code. Eg: onecycle50, step50, gamma')),
'init': scfg.Value('kaiming_normal', help='How to initialized weights. (can be a path to a pretrained model)'),
'pretrained': scfg.Path(help=('alternative way to specify a path to a pretrained model')),
}
def normalize(self):
if self['pretrained'] in ['null', 'None']:
self['pretrained'] = None
if self['pretrained'] is not None:
self['init'] = 'pretrained'
