def from_config(cls, cfg, is_train: bool = True):
augs = utils.build_augmentation(cfg, is_train)
if cfg.INPUT.CROP.ENABLED and is_train:
augs.insert(0, T.RandomCrop(cfg.INPUT.CROP.TYPE, cfg.INPUT.CROP.SIZE))
recompute_boxes = cfg.MODEL.MASK_ON
else:
recompute_boxes = False
ret = {
"is_train": is_train,
"augmentations": augs,
"image_format": cfg.INPUT.FORMAT,
"use_instance_mask": cfg.MODEL.MASK_ON,
"instance_mask_format": cfg.INPUT.MASK_FORMAT,
"use_keypoint": cfg.MODEL.KEYPOINT_ON,
"recompute_boxes": recompute_boxes,
}
if cfg.MODEL.KEYPOINT_ON:
ret["keypoint_hflip_indices"] = utils.create_keypoint_hflip_indices(cfg.DATASETS.TRAIN)
if cfg.MODEL.LOAD_PROPOSALS:
ret["precomputed_proposal_topk"] = (
cfg.DATASETS.PRECOMPUTED_PROPOSAL_TOPK_TRAIN
if is_train
else cfg.DATASETS.PRECOMPUTED_PROPOSAL_TOPK_TEST
)
return ret
def test_flip_keypoints(self):
transforms = T.TransformList([T.HFlipTransform(400)])
anno = {
"bbox": np.asarray([10, 10, 200, 300]),
"bbox_mode": BoxMode.XYXY_ABS,
"keypoints": np.random.rand(17, 3) * 50 + 15,
}
output = detection_utils.transform_instance_annotations(
copy.deepcopy(anno),
transforms,
(400, 400),
keypoint_hflip_indices=detection_utils.create_keypoint_hflip_indices(
["keypoints_coco_2017_train"]
),
)
# The first keypoint is nose
self.assertTrue(np.allclose(output["keypoints"][0, 0], 400 - anno["keypoints"][0, 0]))
# The last 16 keypoints are 8 left-right pairs
self.assertTrue(
np.allclose(
output["keypoints"][1:, 0].reshape(-1, 2)[:, ::-1],
400 - anno["keypoints"][1:, 0].reshape(-1, 2),
)
)
self.assertTrue(
np.allclose(
output["keypoints"][1:, 1:].reshape(-1, 2, 2)[:, ::-1, :],
anno["keypoints"][1:, 1:].reshape(-1, 2, 2),
)
)
def __init__(self, cfg, is_train=True):
self.augmentation = utils.build_augmentation(cfg, is_train)
if cfg.INPUT.CROP.ENABLED and is_train:
self.augmentation.insert(
0, T.RandomCrop(cfg.INPUT.CROP.TYPE, cfg.INPUT.CROP.SIZE))
logging.getLogger(__name__).info("Cropping used in training: " +
str(self.augmentation[0]))
self.compute_tight_boxes = True
else:
self.compute_tight_boxes = False
# fmt: off
self.img_format = cfg.INPUT.FORMAT
self.mask_on = cfg.MODEL.MASK_ON
self.mask_format = cfg.INPUT.MASK_FORMAT
self.keypoint_on = cfg.MODEL.KEYPOINT_ON
self.load_proposals = cfg.MODEL.LOAD_PROPOSALS
# fmt: on
if self.keypoint_on and is_train:
# Flip only makes sense in training
self.keypoint_hflip_indices = utils.create_keypoint_hflip_indices(
cfg.DATASETS.TRAIN)
else:
self.keypoint_hflip_indices = None
if self.load_proposals:
self.proposal_min_box_size = cfg.MODEL.PROPOSAL_GENERATOR.MIN_SIZE
self.proposal_topk = (cfg.DATASETS.PRECOMPUTED_PROPOSAL_TOPK_TRAIN
if is_train else
cfg.DATASETS.PRECOMPUTED_PROPOSAL_TOPK_TEST)
self.is_train = is_train
def __init__(self, cfg, is_train=True):
if cfg.INPUT.CROP.ENABLED:
self.crop_gen = T.RandomCrop(cfg.INPUT.CROP.TYPE, cfg.INPUT.CROP.SIZE)
logging.getLogger(__name__).info("CropGen used in training: " + str(self.crop_gen))
else:
self.crop_gen = None
self.tfm_gens = utils.build_transform_gen(cfg, is_train)
# fmt: off
self.img_format = cfg.INPUT.FORMAT
self.mask_on = cfg.MODEL.MASK_ON
self.mask_format = cfg.INPUT.MASK_FORMAT
self.keypoint_on = cfg.MODEL.KEYPOINT_ON
self.load_proposals = cfg.MODEL.LOAD_PROPOSALS
# fmt: on
if self.keypoint_on and is_train:
# Flip only makes sense in training
self.keypoint_hflip_indices = utils.create_keypoint_hflip_indices(cfg.DATASETS.TRAIN)
else:
self.keypoint_hflip_indices = None
if self.load_proposals:
self.min_box_side_len = cfg.MODEL.PROPOSAL_GENERATOR.MIN_SIZE
self.proposal_topk = (
cfg.DATASETS.PRECOMPUTED_PROPOSAL_TOPK_TRAIN
if is_train
else cfg.DATASETS.PRECOMPUTED_PROPOSAL_TOPK_TEST
)
self.is_train = is_train
def __init__(self, cfg, is_train=True):
self.tfm_gens = utils.build_transform_gen(cfg, is_train)
# fmt: off
self.img_format = cfg.INPUT.FORMAT
self.mask_on = cfg.MODEL.MASK_ON
self.keypoint_on = cfg.MODEL.KEYPOINT_ON
self.densepose_on = cfg.MODEL.DENSEPOSE_ON
assert not cfg.MODEL.LOAD_PROPOSALS, "not supported yet"
# fmt: on
if self.keypoint_on and is_train:
# Flip only makes sense in training
self.keypoint_hflip_indices = utils.create_keypoint_hflip_indices(
cfg.DATASETS.TRAIN)
else:
self.keypoint_hflip_indices = None
if self.densepose_on:
densepose_transform_srcs = [
MetadataCatalog.get(ds).densepose_transform_src
for ds in cfg.DATASETS.TRAIN + cfg.DATASETS.TEST
]
assert len(densepose_transform_srcs) > 0
# TODO: check that DensePose transformation data is the same for
# all the datasets. Otherwise one would have to pass DB ID with
# each entry to select proper transformation data. For now, since
# all DensePose annotated data uses the same data semantics, we
# omit this check.
densepose_transform_data_fpath = PathManager.get_local_path(
densepose_transform_srcs[0])
self.densepose_transform_data = DensePoseTransformData.load(
densepose_transform_data_fpath)
self.is_train = is_train
def __init__(self, cfg, is_train=True):
if cfg.INPUT.CROP.ENABLED and is_train:
self.crop_gen = T.RandomCrop(cfg.INPUT.CROP.TYPE,
cfg.INPUT.CROP.SIZE)
logging.getLogger(__name__).info("CropGen used in training: " +
str(self.crop_gen))
else:
self.crop_gen = None
if is_train:
min_size = cfg.INPUT.MIN_SIZE_TRAIN
max_size = cfg.INPUT.MAX_SIZE_TRAIN
sample_style = cfg.INPUT.MIN_SIZE_TRAIN_SAMPLING
else:
min_size = cfg.INPUT.MIN_SIZE_TEST
max_size = cfg.INPUT.MAX_SIZE_TEST
sample_style = "choice"
if sample_style == "range":
assert len(
min_size
) == 2, "more than 2 ({}) min_size(s) are provided for ranges".format(
len(min_size))
logger = logging.getLogger(__name__)
self.tfm_gens = []
self.tfm_gens.append(
T.ResizeShortestEdge(min_size, max_size, sample_style))
# if self.is_train:
# self.tfm_gens.append(T.RandomBrightness())
# self.tfm_gens.append(T.RandomContrast())
# self.tfm_gens.append(T.RandomLighting())
# self.tfm_gens.append(T.RandomSaturation())
# fmt: off
self.img_format = cfg.INPUT.FORMAT
self.mask_on = cfg.MODEL.MASK_ON
self.mask_format = cfg.INPUT.MASK_FORMAT
self.keypoint_on = cfg.MODEL.KEYPOINT_ON
self.load_proposals = cfg.MODEL.LOAD_PROPOSALS
# fmt: on
if self.keypoint_on and is_train:
# Flip only makes sense in training
self.keypoint_hflip_indices = utils.create_keypoint_hflip_indices(
cfg.DATASETS.TRAIN)
else:
self.keypoint_hflip_indices = None
if self.load_proposals:
self.min_box_side_len = cfg.MODEL.PROPOSAL_GENERATOR.MIN_SIZE
self.proposal_topk = (cfg.DATASETS.PRECOMPUTED_PROPOSAL_TOPK_TRAIN
if is_train else
cfg.DATASETS.PRECOMPUTED_PROPOSAL_TOPK_TEST)
self.is_train = is_train
def __init__(self, cfg, is_train=True):
self.augmentation = build_augmentation(cfg, is_train)
# fmt: off
self.img_format = cfg.INPUT.FORMAT
self.mask_on = (
cfg.MODEL.MASK_ON
or (cfg.MODEL.DENSEPOSE_ON
and cfg.MODEL.ROI_DENSEPOSE_HEAD.COARSE_SEGM_TRAINED_BY_MASKS))
self.keypoint_on = cfg.MODEL.KEYPOINT_ON
self.densepose_on = cfg.MODEL.DENSEPOSE_ON
assert not cfg.MODEL.LOAD_PROPOSALS, "not supported yet"
# fmt: on
if self.keypoint_on and is_train:
# Flip only makes sense in training
self.keypoint_hflip_indices = utils.create_keypoint_hflip_indices(
cfg.DATASETS.TRAIN)
else:
self.keypoint_hflip_indices = None
if self.densepose_on:
densepose_transform_srcs = [
MetadataCatalog.get(ds).densepose_transform_src
for ds in cfg.DATASETS.TRAIN + cfg.DATASETS.TEST
]
# pdb.set_trace()
# densepose_transform_srcs = []
# for ds in cfg.DATASETS.TRAIN + cfg.DATASETS.TEST:
# try:
# ts = MetadataCatalog.get(ds).densepose_transform_src
# except:
# ts = 'https://dl.fbaipublicfiles.com/densepose/data/UV_symmetry_transforms.mat'
# densepose_transform_srcs.append(ts)
assert len(densepose_transform_srcs) > 0
# TODO: check that DensePose transformation data is the same for
# all the datasets. Otherwise one would have to pass DB ID with
# each entry to select proper transformation data. For now, since
# all DensePose annotated data uses the same data semantics, we
# omit this check.
densepose_transform_data_fpath = PathManager.get_local_path(
densepose_transform_srcs[0])
self.densepose_transform_data = DensePoseTransformData.load(
densepose_transform_data_fpath)
self.is_train = is_train
self.use_gt_ins = cfg.MODEL.CONDINST.IUVHead.GT_INSTANCES
self.mask_out_stride = cfg.MODEL.CONDINST.MASK_OUT_STRIDE
self.use_gt_skeleton = cfg.MODEL.CONDINST.IUVHead.GT_SKELETON
if self.use_gt_skeleton:
self.keypoint_on = True
self.use_aux_body_semantics = cfg.MODEL.CONDINST.AUX_SUPERVISION_BODY_SEMANTICS
self.infer_smooth_frame_num = cfg.MODEL.INFERENCE_SMOOTH_FRAME_NUM
def __init__(self, cfg, is_train=True, image_loader=None, tfm_gens=None):
self.tfm_gens = (tfm_gens if tfm_gens is not None else
utils.build_transform_gen(cfg, is_train))
if cfg.INPUT.CROP.ENABLED and is_train:
self.crop_gen = T.RandomCrop(cfg.INPUT.CROP.TYPE,
cfg.INPUT.CROP.SIZE)
# D2GO NOTE: when INPUT.CROP.ENABLED, don't allow using RandomCropOp
assert all(not isinstance(gen, T.RandomCrop)
for gen in self.tfm_gens)
else:
self.crop_gen = None
# fmt: off
self.img_format = cfg.INPUT.FORMAT # noqa
self.mask_on = cfg.MODEL.MASK_ON # noqa
self.mask_format = cfg.INPUT.MASK_FORMAT # noqa
self.keypoint_on = cfg.MODEL.KEYPOINT_ON # noqa
# fmt: on
if self.keypoint_on and is_train:
# Flip only makes sense in training
self.keypoint_hflip_indices = utils.create_keypoint_hflip_indices(
cfg.DATASETS.TRAIN)
else:
self.keypoint_hflip_indices = None
self.load_proposals = cfg.MODEL.LOAD_PROPOSALS
if self.load_proposals:
self.proposal_min_box_size = cfg.MODEL.PROPOSAL_GENERATOR.MIN_SIZE
self.proposal_topk = (cfg.DATASETS.PRECOMPUTED_PROPOSAL_TOPK_TRAIN
if is_train else
cfg.DATASETS.PRECOMPUTED_PROPOSAL_TOPK_TEST)
self.is_train = is_train
# Setup image loader:
self.image_loader = image_loader
self.backfill_size = cfg.D2GO_DATA.MAPPER.BACKFILL_SIZE
self.retry = cfg.D2GO_DATA.MAPPER.RETRY
self.catch_exception = cfg.D2GO_DATA.MAPPER.CATCH_EXCEPTION
if self.backfill_size:
if cfg.DATALOADER.ASPECT_RATIO_GROUPING:
logger.warning(
"ASPECT_RATIO_GROUPING may not work if image's width & height"
" are not given in json dataset when calling extended_coco_load,"
" if you encounter issue, consider disable ASPECT_RATIO_GROUPING."
)
self._error_count = 0
self._total_counts = 0
self._error_types = {}
def __init__(self, cfg, is_train=True):
if cfg.INPUT.CROP.ENABLED and is_train:
# self.crop_gen = T.RandomCrop(cfg.INPUT.CROP.TYPE, cfg.INPUT.CROP.SIZE)
self.crop_gen = T.WiderFace_RandomCrop()
logging.getLogger(__name__).info("CropGen used in training: " +
str(self.crop_gen))
else:
self.crop_gen = None
if is_train:
min_size = cfg.INPUT.MIN_SIZE_TRAIN
max_size = cfg.INPUT.MAX_SIZE_TRAIN
sample_style = cfg.INPUT.MIN_SIZE_TRAIN_SAMPLING
self.tfm_gens = []
# self.tfm_gens.append(T.WiderFace_ResizeShortestEdge(min_size, max_size, sample_style))
self.tfm_gens.append(
T.ResizeShortestEdge(min_size, max_size, sample_style))
self.tfm_gens.append(
T.RandomFlip(prob=0.5, horizontal=True, vertical=False))
self.tfm_gens.append(
T.RandomFlip(prob=0.5, horizontal=False, vertical=True))
# self.tfm_gens.append(T.RandomContrast(0.7, 3.2))
# self.tfm_gens.append(T.RandomBrightness(0.6, 1.8))
# self.tfm_gens.append(T.RandomSaturation(0.6, 1.4))
# self.tfm_gens.append(T.RandomLighting(0.1))
logging.getLogger(__name__).info(
"TransformGens used in training: " + str(self.tfm_gens))
else:
self.tfm_gens = []
self.tfm_gens.append(T.WiderFace_NoOpTransform())
# fmt: off
self.img_format = cfg.INPUT.FORMAT
self.mask_on = cfg.MODEL.MASK_ON
self.mask_format = cfg.INPUT.MASK_FORMAT
self.keypoint_on = cfg.MODEL.KEYPOINT_ON
self.load_proposals = cfg.MODEL.LOAD_PROPOSALS
# fmt: on
if self.keypoint_on and is_train:
# Flip only makes sense in training
self.keypoint_hflip_indices = utils.create_keypoint_hflip_indices(
cfg.DATASETS.TRAIN)
else:
self.keypoint_hflip_indices = None
if self.load_proposals:
self.min_box_side_len = cfg.MODEL.PROPOSAL_GENERATOR.MIN_SIZE
self.proposal_topk = (cfg.DATASETS.PRECOMPUTED_PROPOSAL_TOPK_TRAIN
if is_train else
cfg.DATASETS.PRECOMPUTED_PROPOSAL_TOPK_TEST)
self.is_train = is_train
def __init__(self, cfg, is_train=True):
if cfg.INPUT.CROP.ENABLED and is_train:
self.crop_gen = T.RandomCrop(cfg.INPUT.CROP.TYPE,
cfg.INPUT.CROP.SIZE)
logging.getLogger(__name__).info("CropGen used in training: " +
str(self.crop_gen))
else:
self.crop_gen = None
self.tfm_gens = utils.build_transform_gen(cfg, is_train)
# fmt: off
self.img_format = cfg.INPUT.FORMAT
self.mask_on = cfg.MODEL.MASK_ON
self.mask_format = cfg.INPUT.MASK_FORMAT
self.keypoint_on = cfg.MODEL.KEYPOINT_ON
self.load_proposals = cfg.MODEL.LOAD_PROPOSALS
self.few_shot = cfg.INPUT.FS.FEW_SHOT
self.support_way = cfg.INPUT.FS.SUPPORT_WAY
self.support_shot = cfg.INPUT.FS.SUPPORT_SHOT
# fmt: on
if self.keypoint_on and is_train:
# Flip only makes sense in training
self.keypoint_hflip_indices = utils.create_keypoint_hflip_indices(
cfg.DATASETS.TRAIN)
else:
self.keypoint_hflip_indices = None
if self.load_proposals:
self.proposal_min_box_size = cfg.MODEL.PROPOSAL_GENERATOR.MIN_SIZE
self.proposal_topk = (cfg.DATASETS.PRECOMPUTED_PROPOSAL_TOPK_TRAIN
if is_train else
cfg.DATASETS.PRECOMPUTED_PROPOSAL_TOPK_TEST)
self.is_train = is_train
if self.is_train:
# support_df
self.support_on = True
if self.few_shot:
self.support_df = pd.read_pickle(
"./datasets/coco/10_shot_support_df.pkl")
else:
self.support_df = pd.read_pickle(
"./datasets/coco/train_support_df.pkl")
metadata = MetadataCatalog.get('coco_2017_train')
# unmap the category mapping ids for COCO
reverse_id_mapper = lambda dataset_id: metadata.thing_dataset_id_to_contiguous_id[
dataset_id] # noqa
self.support_df['category_id'] = self.support_df[
'category_id'].map(reverse_id_mapper)
def __init__(self, cfg, is_train=True):
if cfg.INPUT.CROP.ENABLED and is_train:
self.crop_gen = T.RandomCrop(cfg.INPUT.CROP.TYPE,
cfg.INPUT.CROP.SIZE)
logging.getLogger(__name__).info("CropGen used in training: " +
str(self.crop_gen))
else:
self.crop_gen = None
self.tfm_gens = utils.build_transform_gen(cfg, is_train)
# fmt: off
self.img_format = cfg.INPUT.FORMAT
self.mask_on = cfg.MODEL.MASK_ON
self.mask_format = cfg.INPUT.MASK_FORMAT
self.keypoint_on = cfg.MODEL.KEYPOINT_ON
self.load_proposals = cfg.MODEL.LOAD_PROPOSALS
# fmt: on
if self.keypoint_on and is_train:
# Flip only makes sense in training
self.keypoint_hflip_indices = utils.create_keypoint_hflip_indices(
cfg.DATASETS.TRAIN)
else:
self.keypoint_hflip_indices = None
if self.load_proposals:
self.min_box_side_len = cfg.MODEL.PROPOSAL_GENERATOR.MIN_SIZE
self.proposal_topk = (cfg.DATASETS.PRECOMPUTED_PROPOSAL_TOPK_TRAIN
if is_train else
cfg.DATASETS.PRECOMPUTED_PROPOSAL_TOPK_TEST)
self.is_train = is_train
self.vision_type = cfg.INPUT.VISION_TYPE
self.manipulation_type = cfg.INPUT.INFER_MANIPULATION_TYPE
self.manipulation_fun = None
if self.manipulation_type != "":
supported_manipulations = get_testing_augmentations()
self.manipulation_fun = supported_manipulations[
self.manipulation_type]
self.manipulation_value = cfg.INPUT.INFER_MANIPULATION_VALUE
self.opponent_space = cfg.INPUT.OPPONENT_SPACE
self.contrast = cfg.INPUT.CONTRAST
self.mosaic_pattern = cfg.INPUT.MOSAIC_PATTERN
if self.contrast == 1.0:
self.contrast = None
def __init__(self, cfg, is_train=True):
if cfg.INPUT.CROP.ENABLED and is_train:
self.crop_gen = T.RandomCrop(cfg.INPUT.CROP.TYPE,
cfg.INPUT.CROP.SIZE)
logging.getLogger(__name__).info("CropGen used in training: " +
str(self.crop_gen))
else:
self.crop_gen = None
# self.tfm_gens = utils.build_transform_gen(cfg, is_train)
self.tfm_gens = [
T.RandomBrightness(0.1, 1.6),
T.RandomContrast(0.1, 3),
T.RandomSaturation(0.1, 2),
T.RandomRotation(angle=[90, 90]),
T.RandomFlip(prob=0.4, horizontal=False, vertical=True),
T.RandomCrop('relative_range', (0.4, 0.6)),
CutMix()
]
# fmt: off
self.img_format = cfg.INPUT.FORMAT
self.mask_on = cfg.MODEL.MASK_ON
self.mask_format = cfg.INPUT.MASK_FORMAT
self.keypoint_on = cfg.MODEL.KEYPOINT_ON
self.load_proposals = cfg.MODEL.LOAD_PROPOSALS
# fmt: on
if self.keypoint_on and is_train:
# Flip only makes sense in training
self.keypoint_hflip_indices = utils.create_keypoint_hflip_indices(
cfg.DATASETS.TRAIN)
else:
self.keypoint_hflip_indices = None
if self.load_proposals:
self.min_box_side_len = cfg.MODEL.PROPOSAL_GENERATOR.MIN_SIZE
self.proposal_topk = (cfg.DATASETS.PRECOMPUTED_PROPOSAL_TOPK_TRAIN
if is_train else
cfg.DATASETS.PRECOMPUTED_PROPOSAL_TOPK_TEST)
self.is_train = is_train
def __init__(self, cfg, is_train=True):
if cfg.INPUT.CROP.ENABLED and is_train:
self.crop_gen = T.RandomCrop(cfg.INPUT.CROP.TYPE,
cfg.INPUT.CROP.SIZE)
logging.getLogger('detectron2').info("CropGen used in training: " +
str(self.crop_gen))
else:
self.crop_gen = None
self.eval_with_gt = cfg.TEST.get("WITH_GT", False)
self.tfm_gens = build_transform_gen(cfg, is_train)
# fmt: off
self.img_format = cfg.INPUT.FORMAT
self.mask_on = cfg.MODEL.MASK_ON
self.mask_format = cfg.INPUT.MASK_FORMAT
self.keypoint_on = cfg.MODEL.KEYPOINT_ON
self.load_proposals = cfg.MODEL.LOAD_PROPOSALS
# fmt: on
if self.keypoint_on and is_train:
# Flip only makes sense in training
self.keypoint_hflip_indices = utils.create_keypoint_hflip_indices(
cfg.DATASETS.TRAIN)
else:
self.keypoint_hflip_indices = None
if self.load_proposals:
self.min_box_side_len = cfg.MODEL.PROPOSAL_GENERATOR.MIN_SIZE
self.proposal_topk = (cfg.DATASETS.PRECOMPUTED_PROPOSAL_TOPK_TRAIN
if is_train else
cfg.DATASETS.PRECOMPUTED_PROPOSAL_TOPK_TEST)
self.is_train = is_train
self.imgaug_prob = cfg.MODEL.CENTERNET.IMGAUG_PROB
self.kd_without_label = cfg.MODEL.CENTERNET.KD.KD_WITHOUT_LABEL
self.BOX_MINSIZE = cfg.MODEL.CENTERNET.BOX_MINSIZE
from detectron2.data.detection_utils import create_keypoint_hflip_indices
from .coco import dataloader
dataloader.train.dataset.min_keypoints = 1
dataloader.train.dataset.names = "keypoints_coco_2017_train"
dataloader.test.dataset.names = "keypoints_coco_2017_val"
dataloader.train.mapper.update(
use_instance_mask=False,
use_keypoint=True,
keypoint_hflip_indices=create_keypoint_hflip_indices(
dataloader.train.dataset.names),
)
