def test_dataset() -> None:
labels = cache("labels", load_labels)("/store/dataset/labels.csv")
annotations = cache("train_annotations",
get_annotations)("/store/dataset/train.csv", labels)
d = Dataset(annotations)
assert len(d) == 142119
def test_dataset() -> None:
raw_audios = load_audios(RAW_TGT_DIR)
dataset = Dataset(raw_audios, (128, 128))
for i in range(4):
x, y = dataset[0]
plot_spectrograms([np.log(i) for i in [x, y]],
f"/store/plot/test-{i}.png")
def test_dataset(window_size: int, expected: int) -> None:
cache = Cache("/store/tmp")
df = cache("load-train", load)("/store/data/train.csv")
d = Dataset(
df[:20],
window_size=window_size,
stride=5,
)
assert len(d) == expected
x, y = d[0]
assert x.shape == (1, window_size)
assert y.shape == (window_size, )
def test_dataset_total_size() -> None:
cache = Cache("/store/tmp")
df = cache("load-train", load)("/store/data/train.csv")[:20]
d = Dataset(
df,
window_size=5,
stride=5,
)
total_size = 0
for i in range(len(d)):
t, _ = d[i]
total_size += t.shape[1]
assert total_size == 20
def test_transform(id: str, mode: Mode) -> None:
annotations = [Annotation(id, [])]
dataset = Dataset(annotations, mode=mode, resolution=320)
save_image(
make_grid(
[dataset[0][0] for i in range(24)],
nrow=8,
padding=2,
normalize=False,
range=None,
scale_each=False,
pad_value=0,
),
f"/store/tmp/test_aug_{id}-{mode}.png",
)
def test_generation():
dataset = Dataset()
loader = torch.utils.data.DataLoader(dataset, batch_size=1, num_workers=2,
collate_fn=lambda batch: batch)
b0 = None
b1 = None
for i, batch in enumerate(loader):
if i == 0:
b0 = batch
else:
b1 = batch
break
assert b0 != b1
for i, batch in enumerate(loader):
example = batch[0]
ast.parse(example.supervisions["ground_truth"])
if i == 10000:
break
from app.dataset import Dataset
from app.output import Output
from app.segmenter import Segmenter
from app.utils import Utils
from app.debug import Debug
import numpy as np
import cv2
dataset = Dataset()
dataset.load()
segmenter = Segmenter()
Output.clear()
for item in dataset.items:
print(item.file)
#if item.file != "008.jpg":
# continue
#
# run the segmentation algorithm
#
quad = segmenter.segment(item.img, item.distribution_rect)
#
# draw the result
#
# reference image