def test_classification_dataset_additional_meta():
meta = [{
"c1": False,
"c2": True,
"k": 0
}, {
"c1": True,
"c2": False,
"k": 1
}]
seq = random_ubyte_image_sequence(lazy=True, N=2, meta=meta)
ds = ClassificationDataset(seq,
meta_classes=["c2", "c1"],
meta_attrs=["k"])
for i in range(2):
res = ds[i]
x: t.Tensor = res[0]
y: t.Tensor = res[1]
z: int = res[2]
assert isinstance(y, t.Tensor)
assert len(res) == 3
assert y.dtype == t.float32
if i == 0:
assert y.tolist() == [1.0, 0.0]
assert z == 0
else:
assert y.tolist() == [0.0, 1.0]
assert z == 1
def test_dataset():
seq = random_ubyte_image_sequence(lazy=True, N=3)
ds = Dataset(seq)
assert len(ds) == 3
for i in range(3):
assert ds[i].data == seq[i].data
def test_dataset_with_data_transform():
seq = random_ubyte_image_sequence(lazy=True, N=3)
t = lambda x: x + 1.0
ds = Dataset(seq, data_transform=t)
for i in range(3):
assert_equal(ds[i].data, t(seq[i].data))
def test_dataset_with_meta_transform():
meta = [{"k1": v, "k2": -v} for v in range(3)]
seq = random_ubyte_image_sequence(lazy=True, N=3, meta=meta)
t = lambda x: -x
ds = Dataset(seq, meta_attrs=["k2", "k1"], meta_transforms={"k1": t})
for i in range(3):
_, y, z = ds[i]
assert y == -i
assert z == -i
def test_classification_dataset_feed_result():
meta = [{"c1": False, "c2": True}, {"c1": True, "c2": False}]
seq = random_ubyte_image_sequence(lazy=True, N=2, meta=meta)
ds = ClassificationDataset(seq, meta_classes=["c2", "c1"])
result = [
t.tensor([False, True], dtype=t.bool),
t.tensor([True, False], dtype=t.bool),
]
seq2 = ds.result_sequence(result, prefix="xx_")
assert seq2[0].get_meta("xx_c2") == False
assert seq2[0].get_meta("xx_c1") == True
assert seq2[1].get_meta("xx_c2") == True
assert seq2[1].get_meta("xx_c1") == False
def test_dataset_with_meta():
meta = [{"k1": v, "k2": -v} for v in range(3)]
seq = random_ubyte_image_sequence(lazy=True, N=3, meta=meta)
# only use a single meta attr
ds = Dataset(seq, meta_attrs=["k1"])
for i in range(3):
x, y = ds[i]
assert x.data == seq[i].data
assert y == i
# or both
ds = Dataset(seq, meta_attrs=["k2", "k1"])
for i in range(3):
x, y, z = ds[i]
assert x.data == seq[i].data
assert y == -i
assert z == i
