def make_vgg(cfg_id,
batch_norm,
pretrained,
url_id,
incl_fcs=True,
num_classes=1000):
model = VGG(cfgs[cfg_id],
batch_norm=batch_norm,
incl_fcs=incl_fcs,
num_classes=num_classes)
if pretrained:
pretrained_model = model_zoo.load_url(model_urls[url_id])
for k, v in fc_mapping.items():
pretrained_model[k] = pretrained_model.pop(v)
if num_classes != 1000:
del pretrained_model['classifier.weight']
del pretrained_model['classifier.bias']
try:
load_state_dict(model, pretrained_model)
except KeyError:
from jacinle.logging import get_logger
logger = get_logger(__file__)
logger.exception('test')
pass # Intentionally ignore the key error.
return model
def reset_global_seed(seed=None, verbose=False):
if seed is None:
seed = gen_seed()
for k, seed_getter in global_rng_registry.items():
if verbose:
from jacinle.logging import get_logger
logger = get_logger(__file__)
logger.critical('Reset random seed for: {} (pid={}, seed={}).'.format(k, os.getpid(), seed))
seed_getter()(seed)
def git_guard(force=False):
uncommitted_files = git_uncommitted_files()
if len(uncommitted_files) > 0:
from jacinle.logging import get_logger
from jacinle.cli.keyboard import yes_or_no
logger = get_logger(__file__)
logger.warning('Uncommited changes at the current repo:\n ' +
'\n '.join(uncommitted_files))
if force:
if not yes_or_no('Are you sure you want to continue?',
default='no'):
exit(1)
logger.info(git_status_full())
def __init__(self,
counters=None,
display_names=None,
interval=1,
printf=None):
if counters is None:
counters = ['DEFAULT']
self._display_names = display_names
self._counters = collections.OrderedDict([(n, Counter())
for n in counters])
self._interval = interval
self._printf = printf
if self._printf is None:
from jacinle.logging import get_logger
logger = get_logger(__file__)
self._printf = logger.info
import torch
from nscl.datasets.definition import gdef
from nscl.datasets.common.filterable import (
FilterableDatasetUnwrapped,
FilterableDatasetView,
)
from nscl.datasets.common.vocab import Vocab
from nscl.datasets.common.program_translator import (
nsclseq_to_nscltree,
nsclseq_to_nsclqsseq,
nscltree_to_nsclqstree,
gen_vocab,
)
logger = get_logger(__file__)
__all__ = [
"NSCLDataset", "ConceptRetrievalDataset", "ConceptQuantizationDataset"
]
class NSCLDatasetUnwrapped(FilterableDatasetUnwrapped):
def __init__(
self,
scenes_json,
questions_json,
image_root,
image_transform,
vocab_json,
question_transform=None,
