def __init__(self, config, mode="all"):
assert mode in ["train", "validation", "all"
], f"Should be train, validation or all, got {mode}"
self.config = config
self.sequence_length = 30
self.sc = SequenceDataset(MPII_Sequence(config),
self.sequence_length,
step=config["sequence_step_size"])
self.train = int(config["train_size"] * len(self.sc))
self.test = 1 - self.train
self.sigma = config["sigma"]
self.augmentation = config["augmentation"]
self.aug_factor = 0.5
self.resize = iaa.Resize(self.config["resize_to"])
if self.augmentation:
self.seq = iaa.Sequential(
[
# iaa.Sometimes(self.aug_factor, iaa.AdditiveGaussianNoise(scale=0.05 * 255)),
# iaa.Sometimes(self.aug_factor, iaa.SaltAndPepper(0.01, per_channel=False)),
# iaa.Sometimes(self.aug_factor, iaa.CoarseDropout(0.01, size_percent=0.5)),
iaa.Fliplr(self.aug_factor),
iaa.Flipud(self.aug_factor),
# iaa.Sometimes(self.aug_factor, iaa.GaussianBlur(sigma=(0, 3.0))),
# iaa.LinearContrast((0.75, 1.5)),
# Convert each image to grayscale and then overlay the
# result with the original with random alpha. I.e. remove
# colors with varying strengths.
# iaa.Grayscale(alpha=(0.0, 1.0)),
],
random_order=True)
if mode != "all":
# split_indices = np.arange(self.train) if mode == "train" else np.arange(self.train + 1, len(self.sc))
dset_indices = np.arange(len(self.sc))
train_indices, test_indices = sklearn.model_selection.train_test_split(
dset_indices,
train_size=float(config["train_size"]),
random_state=int(config["random_state"]))
if mode == "train":
self.data = SubDataset(self.sc, train_indices)
else:
self.data = SubDataset(self.sc, test_indices)
else:
self.data = self.sc
def __init__(self, config, mode="all"):
assert mode in ["train", "validation", "all"
], f"Should be train, validation or all, got {mode}"
self.config = config
self.sequence_length = 2 # if config.get("sequence_length", False) == False else config["sequence_length"]
# self.sc = Animal_Sequence(config)
self.sc = MPII_Sequence(config)
# works if dataroot like "VOC2011/cats_meta"
# TODO PROBABLY NOT CORRECT HERE
self.animal = config["dataroot"].split("/")[1].split("_")[0]
self.train = int(config["train_size"] * len(self.sc))
self.test = 1 - self.train
self.sigma = config["sigma"]
self.augmentation = config["augmentation"]
self.logger = get_logger(self)
self.resize = iaa.Resize(self.config["resize_to"])
self.aug_factor = 0.5
self.seq = iaa.Sequential([
iaa.Sometimes(self.aug_factor + 0.2, iaa.Fliplr()),
iaa.Sometimes(self.aug_factor, iaa.Flipud()),
],
random_order=True)
if mode != "all":
# split_indices = np.arange(self.train) if mode == "train" else np.arange(self.train + 1, len(self.sc))
dset_indices = np.arange(len(self.sc))
train_indices, test_indices = sklearn.model_selection.train_test_split(
dset_indices,
train_size=float(config["train_size"]),
random_state=int(config["random_state"]))
if mode == "train":
self.data = SubDataset(self.sc, train_indices)
else:
self.data = SubDataset(self.sc, test_indices)
else:
self.data = self.sc
def __init__(self, config, mode="all"):
#super().__init__(config)
assert mode in ["train", "validation", "all"], f"Should be train, validatiopn or all, got {mode}"
self.sc = SingleCats(config)
self.train = int(0.8 * len(self.sc))
self.test = 1 - self.train
self.sigma = self.sc.config["sigma"]
if mode != "all":
# TODO Better split e.g. split per video!
split_indices = np.arange(self.train) if mode == "train" else np.arange(self.train+1, len(self.sc))
self.data = SubDataset(self.sc, split_indices)
else:
self.data = self.sc
def test_sub():
D = DebugDataset(10)
I = np.array([9, 1, 2, 4, 3, 5, 7, 6, 8, 0])
S = SubDataset(D, I)
ref0 = {"val": 9, "other": 9, "index_": 0}
ref2 = {"val": 2, "other": 2, "index_": 2}
ref6 = {"val": 7, "other": 7, "index_": 6}
assert S[0] == ref0
assert S[2] == ref2
assert S[6] == ref6
assert all(S.labels["label1"] == I)
def __init__(self, config, mode="all"):
assert mode in ["train", "validation", "all"], f"Should be train, validation or all, got {mode}"
self.config = config
self.sc = AnimalVOC2011(config)
self.train = int(config["train_size"] * len(self.sc))
self.test = 1 - self.train
self.sigma = config["sigma"]
self.augmentation = config["augmentation"]
self.aug_factor = 0.5
self.resize = iaa.Resize(self.config["resize_to"])
if self.augmentation:
self.seq = iaa.Sequential([
# iaa.Sometimes(self.aug_factor, iaa.AdditiveGaussianNoise(scale=0.05 * 255)),
iaa.Sometimes(self.aug_factor, iaa.SaltAndPepper(0.01, per_channel=False)),
iaa.Sometimes(self.aug_factor, iaa.CoarseDropout(0.01, size_percent=0.5)),
iaa.Fliplr(self.aug_factor),
iaa.Flipud(self.aug_factor),
iaa.Sometimes(self.aug_factor, iaa.GaussianBlur(sigma=(0, 3.0))),
iaa.LinearContrast((0.75, 1.5)),
# Convert each image to grayscale and then overlay the
# result with the original with random alpha. I.e. remove
# colors with varying strengths.
iaa.Grayscale(alpha=(0.0, 1.0)),
], random_order=True)
self.joints = [
[2, 0], # Nose - L_Eye
[2, 1], # Nose - R_Eye
[0, 3], # L_Eye - L_EarBase
[1, 4], # R_Eye - R_EarBase
[2, 8], # Nose - Throat
[8, 9], # Throat - L_F_Elbow
[8, 5], # Throat - R_F_Elbow
[9, 16], # L_F_Elbow - L_F_Knee
[16, 6], # L_F_Knee - L_F_Paw
[5, 17], # R_F_Elbow - R_F_Knee
[17, 7], # R_F_Knee - R_F_Paw
[14, 18], # L_B_Elbow - L_B_Knee
[18, 13], # L_B_Knee - L_B_Paw
[15, 19], # R_B_Elbow - R_B_Knee
[19, 13], # R_B_Knee - R_B_Paw
[10, 11], # Withers - TailBase
]
if self.config.get("min_kps", 0) >= 0:
good_indices = []
for i, entry in enumerate(self.sc):
if len(entry.get("labels_").get("kps")[:, 0].nonzero()[0]) >= self.config.get("min_kps", 0):
good_indices.append(i)
self.sc = SubDataset(self.sc, good_indices).data
if mode != "all":
# split_indices = np.arange(self.train) if mode == "train" else np.arange(self.train + 1, len(self.sc))
dset_indices = np.arange(len(self.sc))
train_indices, test_indices = sklearn.model_selection.train_test_split(dset_indices,
train_size=float(
config["train_size"]),
random_state=int(
config["random_state"]))
if mode == "train":
self.data = SubDataset(self.sc, train_indices)
else:
self.data = SubDataset(self.sc, test_indices)
else:
self.data = self.sc
def labels(data, i):
return {"fid": i}
D = ExtraLabelsDataset(D, labels)
print("D")
for k, v in D.labels.items():
print(k)
print(np.shape(v))
S = SequenceDataset(D, 2)
print("S")
for k, v in S.labels.items():
print(k)
print(np.shape(v))
S = SubDataset(S, [2, 5, 10])
print("Sub")
for k, v in S.labels.items():
print(k)
print(np.shape(v))
U = UnSequenceDataset(S)
print("U")
for k, v in U.labels.items():
print(k)
print(np.shape(v))
print(len(S))
print(U.seq_len)
print(len(U))
def __init__(self, config):
self.P = Prjoti_VUNet(config)
self.data = SubDataset(self.P, np.arange(int(0.9 * len(self.P)), len(self.P)))
def __init__(self,
dataset,
length,
step=1,
fid_key="fid",
strategy="raise"):
"""
Parameters
----------
dataset : DatasetMixin
Dataset from which single frame examples are taken.
length : int
Length of the returned sequences in frames.
step : int
Step between returned frames. Must be `>= 1`.
fid_key : str
Key in labels, at which the frame indices can be found.
strategy : str
How to handle bad sequences, i.e. sequences starting with a
:attr:`fid_key` > 0.
- ``raise``: Raise a ``ValueError``
- ``remove``: remove the sequence
- ``reset``: remove the sequence
This dataset will have `len(dataset) - length * step` examples.
"""
self.step = step
self.length = length
frame_ids = np.array(dataset.labels[fid_key])
if frame_ids.ndim != 1 or len(frame_ids) != len(dataset):
raise ValueError(
"Frame ids must be supplied as a sequence of "
"scalars with the same length as the dataset! Here we "
"have np.shape(dataset.labels[{}]) = {}`.".format(
fid_key, np.shape(frame_ids)))
if frame_ids.dtype != np.int:
raise TypeError(
"Frame ids must be supplied as ints, but are {}".format(
frame_ids.dtype))
if frame_ids.dtype != np.int:
raise TypeError(
"Frame ids must be supplied as ints, but are {}".format(
frame_ids.dtype))
# Gradient
diffs = frame_ids[1:] - frame_ids[:-1]
# All indices where the fid is not monotonically growing
idxs = np.array([0] + list(np.where(diffs != 1)[0] + 1))
# Values at these indices
start_fids = frame_ids[idxs]
# Bad starts
badboys = start_fids != 0
if np.any(badboys):
n = sum(badboys)
i_s = "" if n == 1 else "s"
areis = "is" if n == 1 else "are"
id_s = "ex" if n == 1 else "ices"
if strategy == "raise":
raise ValueError(
"Frame id sequences must always start with 0. "
"There {} {} sequence{} starting with the follwing id{}: "
"{} at ind{} {} in the dataset.".format(
areis, n, i_s, i_s, start_fids[badboys], id_s,
idxs[badboys]))
elif strategy == "remove":
idxs_stop = np.array(list(idxs[1:]) + [None])
starts = idxs[badboys]
stops = idxs_stop[badboys]
bad_seq_mask = np.ones(len(dataset), dtype=bool)
for bad_start_idx, bad_stop_idx in zip(starts, stops):
bad_seq_mask[bad_start_idx:bad_stop_idx] = False
good_seq_idxs = np.arange(len(dataset))[bad_seq_mask]
dataset = SubDataset(dataset, good_seq_idxs)
frame_ids = dataset.labels[fid_key]
elif strategy == "reset":
frame_ids = np.copy(frame_ids) # Don't try to override
idxs_stop = np.array(list(idxs[1:]) + [None])
starts = idxs[badboys]
stops = idxs_stop[badboys]
vals = start_fids[badboys]
for val, bad_sa_idx, bad_so_idx in zip(vals, starts, stops):
frame_ids[bad_sa_idx:bad_so_idx] = (
frame_ids[bad_sa_idx:bad_so_idx] - val)
dataset.labels[fid_key] = frame_ids
frame_ids = dataset.labels[fid_key]
else:
raise ValueError("Strategy of SequenceDataset must be one of "
"`raise`, `remove` or `reset` but is "
"{}".format(strategy))
top_indeces = np.where(np.array(frame_ids) >= (length * step - 1))[0]
all_subdatasets = []
base_indices = []
for i in range(length * step):
indeces = top_indeces - i
base_indices += [indeces]
subdset = SubDataset(dataset, indeces)
all_subdatasets += [subdset]
all_subdatasets = all_subdatasets[::-1]
self.data = ExampleConcatenatedDataset(*all_subdatasets)
self.data.set_example_pars(step=self.step)
self.base_indices = np.array(base_indices).transpose(1, 0)[:, ::-1]
def __init__(self, config, mode="all"):
assert mode in ["train", "validation", "all"], f"Should be train, validation or all, got {mode}"
self.config = config
self.sequence_length = 2 # if config.get("sequence_length", False) == False else config["sequence_length"]
self.sc = SequenceDataset(Animal_Sequence(config), self.sequence_length, step=config["sequence_step_size"])
# works if dataroot like "VOC2011/cats_meta"
# TODO PROBABLY NOT CORRECT HERE
self.animal = config["dataroot"].split("/")[1].split("_")[0]
self.train = int(config["train_size"] * len(self.sc))
self.test = 1 - self.train
self.sigma = config["sigma"]
self.augmentation = config["augmentation"]
self.aug_factor = 0.5
self.logger = get_logger(self)
self.resize = iaa.Resize(self.config["resize_to"])
if self.augmentation:
self.seq = iaa.Sequential([
iaa.Sometimes(self.aug_factor, iaa.AdditiveGaussianNoise(scale=0.05 * 255)),
iaa.Sometimes(self.aug_factor, iaa.SaltAndPepper(0.01, per_channel=False)),
iaa.Sometimes(self.aug_factor, iaa.CoarseDropout(0.01, size_percent=0.5)),
iaa.Fliplr(self.aug_factor),
iaa.Flipud(self.aug_factor),
iaa.Sometimes(self.aug_factor, iaa.GaussianBlur(sigma=(0, 3.0))),
iaa.LinearContrast((0.75, 1.5)),
# Convert each image to grayscale and then overlay the
# result with the original with random alpha. I.e. remove
# colors with varying strengths.
iaa.Grayscale(alpha=(0.0, 1.0)),
], random_order=True)
self.joints = [
[2, 0], # Nose - L_Eye
[2, 1], # Nose - R_Eye
[0, 3], # L_Eye - L_EarBase
[1, 4], # R_Eye - R_EarBase
[2, 8], # Nose - Throat
[8, 9], # Throat - L_F_Elbow
[8, 5], # Throat - R_F_Elbow
[9, 16], # L_F_Elbow - L_F_Knee
[16, 6], # L_F_Knee - L_F_Paw
[5, 17], # R_F_Elbow - R_F_Knee
[17, 7], # R_F_Knee - R_F_Paw
[14, 18], # L_B_Elbow - L_B_Knee
[18, 13], # L_B_Knee - L_B_Paw
[15, 19], # R_B_Elbow - R_B_Knee
[19, 13], # R_B_Knee - R_B_Paw
[10, 11], # Withers - TailBase
]
if mode != "all":
# split_indices = np.arange(self.train) if mode == "train" else np.arange(self.train + 1, len(self.sc))
dset_indices = np.arange(len(self.sc))
train_indices, test_indices = sklearn.model_selection.train_test_split(dset_indices,
train_size=float(
config["train_size"]),
random_state=int(
config["random_state"]))
if mode == "train":
self.data = SubDataset(self.sc, train_indices)
else:
self.data = SubDataset(self.sc, test_indices)
else:
self.data = self.sc