labelme

Semantic Segmentation用のアノテーションツール

Usage

1. Annotate

2. Transform

Scriptを使って変換

code:json_to_png.py

from pathlib import Path

import subprocess

import shutil

import argparse

import tqdm

from PIL import Image

import numpy as np

TEMP_DIR = "temp"

parser = argparse.ArgumentParser()

parser.add_argument("json_dir", help="JSON files directory.")

parser.add_argument("-o", required=True, help="Output directory.")

parser.add_argument("-label_file", required=True, help="Label name file.")

args = parser.parse_args()

json_dir_path = Path(args.json_dir)

out_dir_path = Path(args.o)

label_file_path = Path(args.label_file)

assert json_dir_path.exists(), "JSON directory is not exist."

assert out_dir_path.exists(), "Output directory is not exist."

assert label_file_path.exists(), "Label file is not exist."

# Preprocessing.

correct_label_dict = {}

with open(str(label_file_path), "r") as file:

for i, line in enumerate(file):

temp_dir_path = out_dir_path.joinpath(TEMP_DIR)

if not temp_dir_path.exists():

temp_dir_path.mkdir()

# progress barを出す

for json_file_path in tqdm.tqdm(list(json_dir_path.glob("*.json"))):

stderr=subprocess.PIPE, text=True)

image = Image.open(str(temp_dir_path.joinpath("label.png"))).convert("P")

origin_color_palette = image.getpalette()

image_array = np.array(image)

# Fix label.

label_indexes = {}

with open(str(temp_dir_path.joinpath("label_names.txt")), "r") as label_file:

for label_num, label_name in enumerate(label_file):

label_name = label_name.replace("\n", "")

if label_name == "_background_":

continue

for label_name, label_index in label_indexes.items():

new_image = Image.fromarray(image_array, mode="P")

new_image.putpalette(origin_color_palette)

new_image.save(str(out_dir_path.joinpath(json_file_path.name).with_suffix(".png")))

# Post processing.

if temp_dir_path.exists():

shutil.rmtree(str(temp_dir_path))

print("Done")

python3 json_to_png.py <PATH TO .json> -o <PATH to .png> -label_file=<LABEL NAME>.txt

Misc

Python.icon