def process_all(self, pool: Pool) -> None:
"""Process all images in the folder."""
files = glob.iglob(f"{self.in_dir}/**/*", recursive=True)
images = sorted(filter(utils.has_image_extension, files))
images = [Path(filepath) for filepath in (images)]
utils.tqdm_parallel_imap(self.preprocess, images, pool)
def process_all(self, pool: Pool):
images: List[Path] = []
for root, dirs, files in os.walk(self.in_dir):
root = Path(root)
images.extend([
root / filename for filename in files
if has_image_extension(filename)
])
tqdm_parallel_imap(self.preprocess, images, pool)
def process_all(self, pool: Pool):
"""@TODO: Docs. Contribution is welcome."""
images: List[Path] = []
for root, _, files in os.walk(self.in_dir):
root = Path(root)
images.extend([
root / filename for filename in files
if utils.has_image_extension(filename)
])
utils.tqdm_parallel_imap(self.preprocess, images, pool)
def main(args, _=None):
with get_pool(args.num_workers) as pool:
images = os.listdir(args.in_dir)
colors = tqdm_parallel_imap(colors_in_image, images, pool)
unique_colors = reduce(lambda s1, s2: s1 | s2, colors)
index2color = collections.OrderedDict([
(index, color) for index, color in enumerate(sorted(unique_colors))
])
print("Num classes: ", len(index2color))
with open(args.out_labeling, "w") as fout:
json.dump(index2color, fout, indent=4)
def main(args):
df = pd.read_csv(args.in_csv)
if args.datapath:
df["datapath"] = args.datapath
if args.out_dir:
df["out_dir"] = args.out_dir
df_list = csv2list(df)
with get_pool(args.n_cpu) as pool:
df_list_out = tqdm_parallel_imap(process_row, df_list, pool)
df_out = pd.DataFrame(df_list_out)
df_out.to_csv(args.out_csv, index=False)
def optimize_thresholds(
predictions: np.ndarray,
labels: np.ndarray,
classes: List[int],
metric_fn: Callable = metrics.roc_auc_score,
num_splits: int = 5,
num_repeats: int = 1,
num_workers: int = 0,
ignore_label: int = None,
) -> Tuple[Dict, Dict]:
"""@TODO: Docs. Contribution is welcome."""
pool = utils.get_pool(num_workers)
predictions_copy = predictions.copy()
predictions_list, labels_list = [], []
for class_index in classes:
predictions_list.append(predictions_copy[:, class_index])
labels_list.append(
get_binary_labels(labels, class_index, ignore_label=ignore_label)
)
results = utils.tqdm_parallel_imap(
find_best_threshold_wrapper,
zip(
classes,
predictions_list,
labels_list,
repeat(metric_fn),
repeat(num_splits),
repeat(num_repeats),
),
pool,
)
results = [(r[1], r[2]) for r in sorted(results, key=lambda x: x[0])]
result_thresholds = [r[0] for r in results]
result_metrics = [r[1] for r in results]
class_thresholds = {c: t for (c, t) in zip(classes, result_thresholds)}
class_metrics = {c: m for (c, m) in zip(classes, result_metrics)}
return class_thresholds, class_metrics
def process_all(self, pool: Pool):
images = [*self.in_dir.glob(f"**/*.{self.extension}")]
tqdm_parallel_imap(self.preprocess, images, pool)
def process_all(self, pool: Pool):
images = os.listdir(self.in_dir)
tqdm_parallel_imap(self.preprocess, images, pool)