def build_detection_train_loader(cfg: CfgNode, mapper=None):
"""
A data loader is created in a way similar to that of Detectron2.
The main differences are:
- it allows to combine datasets with different but compatible object category sets
The data loader is created by the following steps:
1. Use the dataset names in config to query :class:`DatasetCatalog`, and obtain a list of dicts.
2. Start workers to work on the dicts. Each worker will:
* Map each metadata dict into another format to be consumed by the model.
* Batch them by simply putting dicts into a list.
The batched ``list[mapped_dict]`` is what this dataloader will return.
Args:
cfg (CfgNode): the config
mapper (callable): a callable which takes a sample (dict) from dataset and
returns the format to be consumed by the model.
By default it will be `DatasetMapper(cfg, True)`.
Returns:
an infinite iterator of training data
"""
_add_category_whitelists_to_metadata(cfg)
_add_category_maps_to_metadata(cfg)
dataset_dicts = combine_detection_dataset_dicts(
cfg.DATASETS.TRAIN,
keep_instance_predicate=_get_train_keep_instance_predicate(cfg),
proposal_files=cfg.DATASETS.PROPOSAL_FILES_TRAIN
if cfg.MODEL.LOAD_PROPOSALS else None,
)
dataset = DatasetFromList(dataset_dicts, copy=False)
if mapper is None:
mapper = DatasetMapper(cfg, True)
dataset = MapDataset(dataset, mapper)
sampler_name = cfg.DATALOADER.SAMPLER_TRAIN
logger = logging.getLogger(__name__)
logger.info("Using training sampler {}".format(sampler_name))
if sampler_name == "TrainingSampler":
sampler = TrainingSampler(len(dataset))
elif sampler_name == "RepeatFactorTrainingSampler":
repeat_factors = RepeatFactorTrainingSampler.repeat_factors_from_category_frequency(
dataset_dicts, cfg.DATALOADER.REPEAT_THRESHOLD)
sampler = RepeatFactorTrainingSampler(repeat_factors)
else:
raise ValueError("Unknown training sampler: {}".format(sampler_name))
return build_batch_data_loader(
dataset,
sampler,
cfg.SOLVER.IMS_PER_BATCH,
aspect_ratio_grouping=cfg.DATALOADER.ASPECT_RATIO_GROUPING,
num_workers=cfg.DATALOADER.NUM_WORKERS,
)
def build_detection_train_loader(cfg, mapper=None):
"""
A data loader is created by the following steps:
1. Use the dataset names in config to query :class:`DatasetCatalog`, and obtain a list of dicts.
2. Coordinate a random shuffle order shared among all processes (all GPUs)
3. Each process spawn another few workers to process the dicts. Each worker will:
* Map each metadata dict into another format to be consumed by the model.
* Batch them by simply putting dicts into a list.
The batched ``list[mapped_dict]`` is what this dataloader will yield.
Args:
cfg (CfgNode): the config
mapper (callable): a callable which takes a sample (dict) from dataset and
returns the format to be consumed by the model.
By default it will be ``DatasetMapper(cfg, True)``.
Returns:
an infinite iterator of training data
"""
dataset_dicts = get_detection_dataset_dicts(
cfg.DATASETS.TRAIN,
filter_empty=cfg.DATALOADER.FILTER_EMPTY_ANNOTATIONS,
min_keypoints=cfg.MODEL.ROI_KEYPOINT_HEAD.MIN_KEYPOINTS_PER_IMAGE
if cfg.MODEL.KEYPOINT_ON
else 0,
proposal_files=cfg.DATASETS.PROPOSAL_FILES_TRAIN if cfg.MODEL.LOAD_PROPOSALS else None,
)
dataset = DatasetFromList(dataset_dicts, copy=False)
if mapper is None:
mapper = DatasetMapper(cfg, True)
dataset = MapDataset(dataset, mapper)
sampler_name = cfg.DATALOADER.SAMPLER_TRAIN
logger = logging.getLogger(__name__)
logger.info("Using training sampler {}".format(sampler_name))
# TODO avoid if-else?
if sampler_name == "TrainingSampler":
sampler = TrainingSampler(len(dataset))
elif sampler_name == "RepeatFactorTrainingSampler":
repeat_factors = RepeatFactorTrainingSampler.repeat_factors_from_category_frequency(
dataset_dicts, cfg.DATALOADER.REPEAT_THRESHOLD
)
sampler = RepeatFactorTrainingSampler(repeat_factors)
else:
raise ValueError("Unknown training sampler: {}".format(sampler_name))
return build_batch_data_loader(
dataset,
sampler,
cfg.SOLVER.IMS_PER_BATCH,
aspect_ratio_grouping=cfg.DATALOADER.ASPECT_RATIO_GROUPING,
num_workers=cfg.DATALOADER.NUM_WORKERS,
)
def build_detection_semisup_train_loader(cfg, mapper=None):
dataset_dicts = get_detection_dataset_dicts(
cfg.DATASETS.TRAIN,
filter_empty=cfg.DATALOADER.FILTER_EMPTY_ANNOTATIONS,
min_keypoints=cfg.MODEL.ROI_KEYPOINT_HEAD.MIN_KEYPOINTS_PER_IMAGE
if cfg.MODEL.KEYPOINT_ON else 0,
proposal_files=cfg.DATASETS.PROPOSAL_FILES_TRAIN
if cfg.MODEL.LOAD_PROPOSALS else None,
)
# Divide into labeled and unlabeled sets according to supervision percentage
label_dicts, unlabel_dicts = divide_label_unlabel(
dataset_dicts,
cfg.DATALOADER.SUP_PERCENT,
cfg.DATALOADER.RANDOM_DATA_SEED,
cfg.DATALOADER.RANDOM_DATA_SEED_PATH,
)
dataset = DatasetFromList(label_dicts, copy=False)
if mapper is None:
mapper = DatasetMapper(cfg, True)
dataset = MapDataset(dataset, mapper)
sampler_name = cfg.DATALOADER.SAMPLER_TRAIN
logger = logging.getLogger(__name__)
logger.info("Using training sampler {}".format(sampler_name))
if sampler_name == "TrainingSampler":
sampler = TrainingSampler(len(dataset))
elif sampler_name == "RepeatFactorTrainingSampler":
repeat_factors = (
RepeatFactorTrainingSampler.repeat_factors_from_category_frequency(
label_dicts, cfg.DATALOADER.REPEAT_THRESHOLD))
sampler = RepeatFactorTrainingSampler(repeat_factors)
else:
raise ValueError("Unknown training sampler: {}".format(sampler_name))
# list num of labeled and unlabeled
logger.info("Number of training samples " + str(len(dataset)))
logger.info("Supervision percentage " + str(cfg.DATALOADER.SUP_PERCENT))
return build_batch_data_loader(
dataset,
sampler,
cfg.SOLVER.IMS_PER_BATCH,
aspect_ratio_grouping=cfg.DATALOADER.ASPECT_RATIO_GROUPING,
num_workers=cfg.DATALOADER.NUM_WORKERS,
)
def test_repeat_factors_from_category_frequency(self):
repeat_thresh = 0.5
dataset_dicts = [
{
"annotations": [{
"category_id": 0
}, {
"category_id": 1
}]
},
{
"annotations": [{
"category_id": 0
}]
},
{
"annotations": []
},
]
rep_factors = RepeatFactorTrainingSampler.repeat_factors_from_category_frequency(
dataset_dicts, repeat_thresh)
expected_rep_factors = torch.tensor([math.sqrt(3 / 2), 1.0, 1.0])
self.assertTrue(torch.allclose(rep_factors, expected_rep_factors))
def build_custom_train_loader(cfg, mapper=None):
"""
Modified from detectron2.data.build.build_custom_train_loader, but supports
different samplers
"""
dataset_dicts = get_detection_dataset_dicts(
cfg.DATASETS.TRAIN,
filter_empty=cfg.DATALOADER.FILTER_EMPTY_ANNOTATIONS,
min_keypoints=cfg.MODEL.ROI_KEYPOINT_HEAD.MIN_KEYPOINTS_PER_IMAGE
if cfg.MODEL.KEYPOINT_ON else 0,
proposal_files=cfg.DATASETS.PROPOSAL_FILES_TRAIN
if cfg.MODEL.LOAD_PROPOSALS else None,
)
dataset = DatasetFromList(dataset_dicts, copy=False)
if mapper is None:
mapper = DatasetMapper(cfg, True)
dataset = MapDataset(dataset, mapper)
sampler_name = cfg.DATALOADER.SAMPLER_TRAIN
logger = logging.getLogger(__name__)
logger.info("Using training sampler {}".format(sampler_name))
# TODO avoid if-else?
if sampler_name == "TrainingSampler":
sampler = TrainingSampler(len(dataset))
elif sampler_name == "RepeatFactorTrainingSampler":
repeat_factors = RepeatFactorTrainingSampler.repeat_factors_from_category_frequency(
dataset_dicts, cfg.DATALOADER.REPEAT_THRESHOLD)
sampler = RepeatFactorTrainingSampler(repeat_factors)
elif sampler_name == "ClassAwareSampler":
sampler = ClassAwareSampler(dataset_dicts)
else:
raise ValueError("Unknown training sampler: {}".format(sampler_name))
return build_batch_data_loader(
dataset,
sampler,
cfg.SOLVER.IMS_PER_BATCH,
aspect_ratio_grouping=cfg.DATALOADER.ASPECT_RATIO_GROUPING,
num_workers=cfg.DATALOADER.NUM_WORKERS,
)
def _train_loader_from_config(cfg, mapper=None, *, dataset=None, sampler=None):
if dataset is None:
dataset = get_detection_dataset_dicts(
cfg.DATASETS.TRAIN,
filter_empty=cfg.DATALOADER.FILTER_EMPTY_ANNOTATIONS,
min_keypoints=cfg.MODEL.ROI_KEYPOINT_HEAD.MIN_KEYPOINTS_PER_IMAGE
if cfg.MODEL.KEYPOINT_ON
else 0,
proposal_files=cfg.DATASETS.PROPOSAL_FILES_TRAIN if cfg.MODEL.LOAD_PROPOSALS else None,
)
if mapper is None:
mapper = DatasetMapper(cfg, True)
if sampler is None:
sampler_name = cfg.DATALOADER.SAMPLER_TRAIN
logger = logging.getLogger("detectron2.trainer")
logger.info("Using training sampler {}".format(sampler_name))
if sampler_name == "TrainingSampler":
sampler = TrainingSampler(len(dataset))
elif sampler_name == "RepeatFactorTrainingSampler":
repeat_factors = RepeatFactorTrainingSampler.repeat_factors_from_category_frequency(
dataset, cfg.DATALOADER.REPEAT_THRESHOLD
)
sampler = RepeatFactorTrainingSampler(repeat_factors)
else:
raise ValueError(
"Unknown training sampler: {}".format(sampler_name))
return {
"dataset": dataset,
"sampler": sampler,
"mapper": mapper,
"total_batch_size": cfg.SOLVER.IMS_PER_BATCH,
"aspect_ratio_grouping": cfg.DATALOADER.ASPECT_RATIO_GROUPING,
"num_workers": cfg.DATALOADER.NUM_WORKERS,
}
def build_weighted_detection_train_loader(cfg: CfgNode, mapper=None):
dataset_repeat_factors = get_train_datasets_repeat_factors(cfg)
# OrderedDict to guarantee order of values() consistent with repeat factors
dataset_name_to_dicts = OrderedDict(
{
name: get_detection_dataset_dicts(
[name],
filter_empty=cfg.DATALOADER.FILTER_EMPTY_ANNOTATIONS,
min_keypoints=cfg.MODEL.ROI_KEYPOINT_HEAD.MIN_KEYPOINTS_PER_IMAGE
if cfg.MODEL.KEYPOINT_ON
else 0,
proposal_files=cfg.DATASETS.PROPOSAL_FILES_TRAIN
if cfg.MODEL.LOAD_PROPOSALS
else None,
)
for name in cfg.DATASETS.TRAIN
}
)
# Repeat factor for every sample in the dataset
repeat_factors = [
[dataset_repeat_factors[dsname]] * len(dataset_name_to_dicts[dsname])
for dsname in cfg.DATASETS.TRAIN
]
repeat_factors = list(itertools.chain.from_iterable(repeat_factors))
dataset_dicts = dataset_name_to_dicts.values()
dataset_dicts = list(itertools.chain.from_iterable(dataset_dicts))
dataset = DatasetFromList(dataset_dicts, copy=False)
if mapper is None:
mapper = DatasetMapper(cfg, True)
dataset = MapDataset(dataset, mapper)
logger.info(
"Using WeightedTrainingSampler with repeat_factors={}".format(
cfg.DATASETS.TRAIN_REPEAT_FACTOR
)
)
sampler = RepeatFactorTrainingSampler(torch.tensor(repeat_factors))
return build_batch_data_loader(
dataset,
sampler,
cfg.SOLVER.IMS_PER_BATCH,
aspect_ratio_grouping=cfg.DATALOADER.ASPECT_RATIO_GROUPING,
num_workers=cfg.DATALOADER.NUM_WORKERS,
)
# %% --------------------
# https://detectron2.readthedocs.io/en/latest/modules/data.html#detectron2.data.samplers.RepeatFactorTrainingSampler
# https://github.com/facebookresearch/detectron2/issues/2139
dataset_dicts_1 = [
{"annotations": [{"category_id": 0}, {"category_id": 1}]},
{"annotations": [{"category_id": 0}]},
{"annotations": [{"category_id": 0}]},
{"annotations": [{"category_id": 0}]},
{"annotations": [{"category_id": 0}, {"category_id": 2}]},
{"annotations": [{"category_id": 0}, {"category_id": 2}]},
{"annotations": [{"category_id": 0}]},
{"annotations": [{"category_id": 0}]},
{"annotations": [{"category_id": 0}]},
{"annotations": [{"category_id": 0}]},
{"annotations": [{"category_id": 0}]},
{"annotations": [{"category_id": 0}]},
{"annotations": [{"category_id": 0}]},
{"annotations": [{"category_id": 0}]},
{"annotations": [{"category_id": 0}]},
{"annotations": [{"category_id": 0}]},
{"annotations": [{"category_id": 0}]},
{"annotations": [{"category_id": 0}]},
{"annotations": [{"category_id": 0}]},
]
sampler = RepeatFactorTrainingSampler.repeat_factors_from_category_frequency(dataset_dicts_1, 3)
# Inherit the Sampler, based on the threshold to reduce the problem of sample imbalance,
# categories with fewer samples are more likely to be sampled, which is suitable for the LVIS
# data set
print(sampler)