def __repr__(self):
if (__COULD_HAVE_IPYTHON__ and hasattr(get_ipython(), "config")
and "IPKernelApp" in get_ipython().config):
label_str = pp2mkdtable(self.labels, True)
else:
label_str = pp2mkdtable(self.labels, False)
descr = self.meta.get("description", "MetaDataset")
repr_str = f"{descr}\n\n# Labels\n{label_str}"
return repr_str
def explore(config, disable_cache=False):
if not disable_cache:
get_state = st.cache(persist=False, allow_output_mutation=True)(_get_state)
else:
get_state = _get_state
dset = get_state(config)
dset.expand = True
st.title("Dataset Explorer: {}".format(type(dset).__name__))
input_method = st.sidebar.selectbox(
"Index selection method", ["Slider", "Number input", "Sample"]
)
if input_method == "Slider":
idx = st.sidebar.slider("Index", 0, len(dset), 0)
elif input_method == "Number input":
idx = st.sidebar.number_input("Index", 0, len(dset), 0)
elif input_method == "Sample":
idx = 0
if st.sidebar.button("Sample"):
idx = np.random.choice(len(dset))
st.sidebar.text("Index: {}".format(idx))
show_example(dset, idx)
st.header("config")
cfg_string = pp2mkdtable(config, jupyter_style=True)
cfg = st.markdown(cfg_string)
def display_status(stages_per_vid, stages=STAGES, error=ERROR):
to_print = {}
finished = []
for vid, completed in stages_per_vid.items():
seen_stages = np.array(completed)[:, 0]
if error in seen_stages:
to_print[vid] = completed[0, 1]
elif stages[-1] in seen_stages:
finished += [vid]
continue
else:
def sort_key(v):
v = v[0]
if ' - skipped' in v:
element = v[:len(v) - len(' - skipped')]
else:
element = v
return stages.index(element)
to_print[vid] = sorted(completed, key=sort_key)[-1][0]
to_print = pp2mkdtable(to_print)
print(to_print)
print('Finished {} videos'.format(len(finished)))
return to_print
def show_example(dset, idx):
ex = dset[idx]
st.header("Keys")
walk(ex, display, pass_key=True)
st.header("Summary")
summary = pp2mkdtable(ex, jupyter_style=True)
# print markdown summary on console for easy copy and pasting in readme etc
print(summary)
st.markdown(summary)
def explore(config, disable_cache=False):
if not disable_cache:
get_state = st.cache(persist=False,
allow_output_mutation=True)(_get_state)
else:
get_state = _get_state
dset = get_state(config)
dset.expand = True
st.title("Dataset Explorer: {}".format(type(dset).__name__))
idx = st.sidebar.slider("index", 0, len(dset), 0)
if st.sidebar.button("sample"):
idx = np.random.choice(len(dset))
show_example(dset, idx)
st.header("config")
cfg_string = pp2mkdtable(config, jupyter_style=True)
cfg = st.markdown(cfg_string)
ex["keypoints_reference"] = [
e["keypoints"][self.c] for e in ex["neighbours"]
]
ex["image_reference"] = [
e["target"][self.c] for e in ex["neighbours"]
]
ex["magnification_factor"] = self.mags
ex["frame_anchor_1"] = ex["image_reference"][0]
ex["frame_anchor_2"] = ex["image_query"]
return ex
def __len__(self):
return len(self.viterbi_neighbours)
return ABC_Seq_Mag_Dset
if __name__ == "__main__":
from abc_interpolation.data.human_gait import HumanGaitFixedBox
dataset = HumanGaitFixedBox
hg_viterbi = make_abc_nn_seq_mag_dset(dataset)
data = hg_viterbi({"data_split": "train", "mode": "eval"})
from edflow.util import pp2mkdtable
pp2mkdtable(data.get_example(10))
debug = False
D = AggregatedMultiPersonDataset(
{'spatial_size': 256},
root='/export/scratch/jhaux/Data/trickyoga',
ext='mp4',
force=True, # See if new videos are ready!
debug=debug)
D2 = AggregatedMultiPersonDataset(
{'spatial_size': 256},
root='/export/scratch/jhaux/Data/olympic_sports_new',
ext='seq',
force=False, # See if new videos are ready!
debug=debug)
d = D[10]
tab = pp2mkdtable(d)
plot_datum(d, 'ty_10.png')
print(tab)
d = D2[10]
tab = pp2mkdtable(d)
plot_datum(d, 'oly_10.png')
print(tab)
print("D1: {}\nD2: {}".format(len(D), len(D2)))
with open('multiperson.md', 'w+') as df:
df.write(tab)
example["image"] = example["image"] / 127.5 - 1.0
example["image"] = example["image"].astype(np.float32)
def get_example(self, i):
example = self._load_example(i)
self._preprocess_example(example)
return example
def __len__(self):
return self._length
if __name__ == "__main__":
from edflow.util import pp2mkdtable
print("train")
d = CIFAR10()
print(len(d))
e = d[0]
print(pp2mkdtable(e))
x, y = e["image"], e["class"]
print(x.dtype, x.shape, x.min(), x.max(), y)
print("test")
d = CIFAR10({"CIFAR10": {"split": "test"}})
print(len(d))
from PIL import Image
Image.fromarray(
((x + 1.0) * 127.5).astype(np.uint8)).save("cifar10_example.png")
d['im_crop'] = im_crop
plot_kps(d, idx)
if __name__ == '__main__':
from multiperson_dataset import MultiPersonDataset
# MP = MultiPersonDataset('/export/scratch/jhaux/Data/olympic sports/')
MP = MultiPersonDataset('/export/scratch/jhaux/Data/olympic_test/')
CMP = CropDataset(MP)
from edflow.util import pprint, pp2mkdtable
from edflow.data.util import plot_datum
idx = 10
for idx in range(25):
d = CMP[idx]
print(pp2mkdtable(d))
plot_datum(d, 'crop_{}.png'.format(idx))
test_new_crop(d, idx)
for idx in np.random.randint(len(CMP), size=10):
d = CMP[idx]
print(pp2mkdtable(d))
plot_datum(d, 'crop_{}.png'.format(idx))
test_new_crop(d, idx)
def show_example(dset, idx):
ex = dset[idx]
st.header("Keys")
walk(ex, display, pass_key=True)
st.header("Summary")
st.markdown(pp2mkdtable(ex, jupyter_style=True))