Files
UTILITY_AI_ANNOTATION_TOOL/mmdet_annotation_modify.py
2025-11-17 18:00:06 +09:00

579 lines
22 KiB
Python

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
COCO Annotation Utility Tool
COCO format annotation file manipulation utility
Usage examples:
# Interactive mode (recommended)
python Utility_lableing_tool.py
# Remove classes
python Utility_lableing_tool.py remove --input annotations.json --output new.json --classes "person,car"
# Merge classes
python Utility_lableing_tool.py merge --input annotations.json --output new.json --source "car,truck,bus" --target "vehicle"
# Rename classes
python Utility_lableing_tool.py rename --input annotations.json --output new.json --mapping "old_name:new_name,another_old:another_new"
# Show info
python Utility_lableing_tool.py info --input annotations.json
"""
import json
import argparse
from pathlib import Path
from typing import Dict, List, Set, Union
import shutil
from datetime import datetime
import sys
import os
class COCOAnnotationEditor:
"""Class for editing COCO format annotations"""
def __init__(self, annotation_path: str):
"""
Args:
annotation_path: Path to COCO format annotation JSON file
"""
self.annotation_path = Path(annotation_path)
self.data = self._load_annotation()
def _load_annotation(self) -> dict:
"""Load annotation file"""
with open(self.annotation_path, 'r', encoding='utf-8') as f:
return json.load(f)
def save(self, output_path: str, backup: bool = True):
"""
Save modified annotation
Args:
output_path: Output file path
backup: Whether to backup original file
"""
output_path = Path(output_path)
# Create backup
if backup and output_path.exists():
backup_path = output_path.parent / f"{output_path.stem}_backup_{datetime.now().strftime('%Y%m%d_%H%M%S')}{output_path.suffix}"
shutil.copy2(output_path, backup_path)
print(f"[+] 백업 생성됨: {backup_path}")
# Save
with open(output_path, 'w', encoding='utf-8') as f:
json.dump(self.data, f, ensure_ascii=False, indent=2)
print(f"[+] 저장 완료: {output_path}")
def get_category_info(self) -> Dict:
"""Get category information"""
categories = self.data.get('categories', [])
annotations = self.data.get('annotations', [])
# Count annotations per category
category_counts = {}
for ann in annotations:
cat_id = ann['category_id']
category_counts[cat_id] = category_counts.get(cat_id, 0) + 1
info = {}
for cat in categories:
cat_id = cat['id']
info[cat_id] = {
'id': cat_id,
'name': cat['name'],
'supercategory': cat.get('supercategory', ''),
'annotation_count': category_counts.get(cat_id, 0)
}
return info
def print_info(self):
"""Print annotation information"""
print("\n" + "="*60)
print(f"파일: {self.annotation_path}")
print("="*60)
print(f"\n전체 통계:")
print(f" - 이미지 수: {len(self.data.get('images', []))}")
print(f" - 어노테이션 수: {len(self.data.get('annotations', []))}")
print(f" - 클래스 수: {len(self.data.get('categories', []))}")
print(f"\n클래스 상세:")
print(f"{'ID':<6} {'클래스명':<30} {'어노테이션 수':<15}")
print("-" * 55)
category_info = self.get_category_info()
total_annotations = 0
for cat_id, info in sorted(category_info.items()):
ann_count = info['annotation_count']
total_annotations += ann_count
print(f"{info['id']:<6} {info['name']:<30} {ann_count:<15}")
print("-" * 55)
print(f"{'합계':<37} {total_annotations:<15}")
print("="*60 + "\n")
def remove_categories(self, category_names: List[str]) -> 'COCOAnnotationEditor':
"""
Remove specific categories
Args:
category_names: List of category names or IDs to remove
"""
# Build ID to name mapping
id_to_cat = {cat['id']: cat for cat in self.data.get('categories', [])}
name_to_cat = {cat['name']: cat for cat in self.data.get('categories', [])}
# Find category IDs to remove
categories_to_remove = set()
for item in category_names:
# Try to parse as ID first
try:
cat_id = int(item)
if cat_id in id_to_cat:
cat = id_to_cat[cat_id]
categories_to_remove.add(cat['id'])
print(f" - 제거: {cat['name']} (ID: {cat['id']})")
else:
print(f"[!] ID {cat_id}를 찾을 수 없습니다")
except ValueError:
# Not a number, treat as name
if item in name_to_cat:
cat = name_to_cat[item]
categories_to_remove.add(cat['id'])
print(f" - 제거: {cat['name']} (ID: {cat['id']})")
else:
print(f"[!] 이름 '{item}'을 찾을 수 없습니다")
# Update categories
remaining_categories = [
cat for cat in self.data.get('categories', [])
if cat['id'] not in categories_to_remove
]
# Update categories
self.data['categories'] = remaining_categories
# Remove annotations of removed categories
original_count = len(self.data.get('annotations', []))
self.data['annotations'] = [
ann for ann in self.data.get('annotations', [])
if ann['category_id'] not in categories_to_remove
]
removed_count = original_count - len(self.data['annotations'])
print(f"[+] {removed_count}개의 어노테이션이 제거되었습니다")
return self
def merge_categories(self, source_names: List[str], target_name: str,
target_supercategory: str = '') -> 'COCOAnnotationEditor':
"""
Merge multiple categories into one
Args:
source_names: List of source category names or IDs to merge
target_name: Target category name after merge
target_supercategory: Target supercategory (optional)
"""
# Find source category IDs
source_ids = set()
source_categories = []
# Build ID to name mapping
id_to_cat = {cat['id']: cat for cat in self.data.get('categories', [])}
name_to_cat = {cat['name']: cat for cat in self.data.get('categories', [])}
for item in source_names:
# Try to parse as ID first
try:
cat_id = int(item)
if cat_id in id_to_cat:
cat = id_to_cat[cat_id]
source_ids.add(cat['id'])
source_categories.append(cat)
print(f" - 통합 대상: {cat['name']} (ID: {cat['id']})")
else:
print(f"[!] ID {cat_id}를 찾을 수 없습니다")
except ValueError:
# Not a number, treat as name
if item in name_to_cat:
cat = name_to_cat[item]
source_ids.add(cat['id'])
source_categories.append(cat)
print(f" - 통합 대상: {cat['name']} (ID: {cat['id']})")
else:
print(f"[!] 이름 '{item}'을 찾을 수 없습니다")
if not source_ids:
print(f"[!] 통합할 클래스를 찾을 수 없습니다")
return self
# New category ID (use first source ID)
new_category_id = min(source_ids)
# Create new category
new_category = {
'id': new_category_id,
'name': target_name,
'supercategory': ''
}
# Remove source categories and add new category
self.data['categories'] = [
cat for cat in self.data.get('categories', [])
if cat['id'] not in source_ids
]
self.data['categories'].append(new_category)
# Update annotations (change all source IDs to new ID)
for ann in self.data.get('annotations', []):
if ann['category_id'] in source_ids:
ann['category_id'] = new_category_id
print(f"[+] '{target_name}' (ID: {new_category_id})로 통합되었습니다")
return self
def rename_categories(self, mapping: Dict[str, str]) -> 'COCOAnnotationEditor':
"""
Rename categories
Args:
mapping: Dictionary of {old_name_or_id: new_name}
"""
# Build ID to cat mapping
id_to_cat = {cat['id']: cat for cat in self.data.get('categories', [])}
name_to_cat = {cat['name']: cat for cat in self.data.get('categories', [])}
renamed_count = 0
for old_key, new_name in mapping.items():
# Try to parse as ID first
try:
cat_id = int(old_key)
if cat_id in id_to_cat:
cat = id_to_cat[cat_id]
old_name = cat['name']
cat['name'] = new_name
print(f" - 변경: '{old_name}''{new_name}' (ID: {cat['id']})")
renamed_count += 1
else:
print(f"[!] ID {cat_id}를 찾을 수 없습니다")
except ValueError:
# Not a number, treat as name
if old_key in name_to_cat:
cat = name_to_cat[old_key]
cat['name'] = new_name
print(f" - 변경: '{old_key}''{new_name}' (ID: {cat['id']})")
renamed_count += 1
else:
print(f"[!] 이름 '{old_key}'를 찾을 수 없습니다")
print(f"[+] {renamed_count}개 클래스 이름이 변경되었습니다")
return self
def reindex_categories(self, start_id: int = 1) -> 'COCOAnnotationEditor':
"""
Reindex category IDs sequentially
Args:
start_id: Starting ID (default: 1)
"""
# Create mapping: old ID -> new ID
id_mapping = {}
new_id = start_id
for cat in sorted(self.data.get('categories', []), key=lambda x: x['id']):
old_id = cat['id']
id_mapping[old_id] = new_id
print(f" - ID 변경: {old_id}{new_id} ({cat['name']})")
new_id += 1
# Update category IDs
for cat in self.data.get('categories', []):
cat['id'] = id_mapping[cat['id']]
# Update annotation category IDs
for ann in self.data.get('annotations', []):
ann['category_id'] = id_mapping[ann['category_id']]
print(f"[+] 클래스 ID가 재할당되었습니다 (시작: {start_id})")
return self
def interactive_mode():
"""Interactive mode for user-friendly editing"""
print("\n" + "="*60)
print(" COCO 어노테이션 편집기 - 대화형 모드")
print("="*60)
# 1. Select annotation file
print("\n[1/3] 어노테이션 파일 선택")
print("-" * 60)
# Find annotation files in common locations
common_paths = [
"dataset/*/annotations/*.json",
"data/*/annotations/*.json",
"annotations/*.json",
"*.json"
]
from glob import glob
found_files = []
for pattern in common_paths:
found_files.extend(glob(pattern, recursive=True))
input_file = None
while not input_file:
if found_files:
print("\n발견된 어노테이션 파일:")
for idx, f in enumerate(found_files[:10], 1):
print(f" {idx}. {f}")
if len(found_files) > 10:
print(f" ... 외 {len(found_files) - 10}개 더")
print(f" 0. 직접 경로 입력")
choice = input("\n파일 번호 선택 (직접 입력은 0): ").strip()
if choice == '0':
input_file = input("어노테이션 파일 경로 입력: ").strip()
else:
try:
input_file = found_files[int(choice) - 1]
except (ValueError, IndexError):
print("[!] 잘못된 선택입니다. 다시 선택해주세요.")
continue
else:
input_file = input("어노테이션 파일 경로 입력: ").strip()
if not Path(input_file).exists():
print(f"[!] 파일을 찾을 수 없습니다: {input_file}")
print("[!] 다시 입력해주세요.")
input_file = None
# Load and show info
print(f"\n[+] 로딩 중: {input_file}")
editor = COCOAnnotationEditor(input_file)
editor.print_info()
# Store original categories for reference
original_categories = {cat['id']: cat['name'] for cat in editor.data.get('categories', [])}
# 2. Select operations
print("\n[2/3] 작업 선택")
print("-" * 60)
operations = []
while True:
print("\n사용 가능한 작업:")
print(" 1. 클래스 제거")
print(" 2. 클래스 통합 (병합)")
print(" 3. 클래스 이름 변경")
print(" 4. 클래스 ID 재할당")
print(" 5. 현재 정보 표시")
print(" 0. 완료 (저장 단계로)")
choice = input("\n작업 선택 (0-5): ").strip()
if choice == '0':
if not operations:
print("[!] 선택된 작업이 없습니다. 종료합니다...")
return
break
elif choice == '1': # Remove
print("\n현재 클래스:")
for cat in editor.data.get('categories', []):
print(f" - {cat['name']} (ID: {cat['id']})")
classes = input("\n제거할 클래스 입력 (이름 또는 ID, 쉼표로 구분): ").strip()
if classes:
class_list = [c.strip() for c in classes.split(',')]
print(f"\n[*] 클래스 제거 중: {class_list}")
editor.remove_categories(class_list)
operations.append(f"제거: {', '.join(class_list)}")
elif choice == '2': # Merge
print("\n현재 클래스:")
for cat in editor.data.get('categories', []):
print(f" - {cat['name']} (ID: {cat['id']})")
source = input("\n통합할 클래스 입력 (이름 또는 ID, 쉼표로 구분): ").strip()
if source:
target = input("통합 후 클래스 이름 입력: ").strip()
source_list = [c.strip() for c in source.split(',')]
print(f"\n[*] '{target}'로 통합 중: {source_list}")
editor.merge_categories(source_list, target, '')
operations.append(f"통합: {', '.join(source_list)}{target}")
elif choice == '3': # Rename
print("\n현재 클래스:")
for cat in editor.data.get('categories', []):
print(f" - {cat['name']} (ID: {cat['id']})")
print("\n매핑 입력 (형식: 기존:새이름,기존:새이름 - 쉼표로 구분)")
print("예시: tt:transformer,ss:substation 또는 1:transformer,2:substation")
mapping_input = input("\n입력: ").strip()
if mapping_input:
mapping = {}
for pair in mapping_input.split(','):
if ':' in pair:
old, new = pair.split(':', 1)
mapping[old.strip()] = new.strip()
if mapping:
print(f"\n[*] 클래스 이름 변경 중")
editor.rename_categories(mapping)
operations.append(f"이름 변경: {len(mapping)}개 클래스")
elif choice == '4': # Reindex
while True:
start_id = input("\n시작 ID 입력 (기본값: 1): ").strip()
if not start_id:
start_id = 1
break
try:
start_id = int(start_id)
break
except ValueError:
print("[!] 숫자를 입력해주세요.")
print(f"\n[*] 클래스 ID 재할당 중 (시작: {start_id})")
editor.reindex_categories(start_id=start_id)
operations.append(f"ID 재할당: {start_id}부터")
elif choice == '5': # Show info
editor.print_info()
else:
print("[!] 잘못된 선택입니다")
# 3. Save
print("\n[3/3] 결과 저장")
print("-" * 60)
print("\n수행된 작업:")
for idx, op in enumerate(operations, 1):
print(f" {idx}. {op}")
print(f"\n입력 파일: {input_file}")
default_output = str(Path(input_file).parent / f"{Path(input_file).stem}_edited{Path(input_file).suffix}")
output_file = input(f"\n출력 파일 경로 입력 (기본값: {default_output}): ").strip()
if not output_file:
output_file = default_output
create_backup = input("출력 파일이 존재하면 백업 생성? (Y/n): ").strip().lower()
backup = create_backup != 'n'
print(f"\n[*] 저장 중: {output_file}...")
editor.save(output_file, backup=backup)
print("\n" + "="*60)
print(" 작업이 성공적으로 완료되었습니다!")
print("="*60)
print(f"\n입력: {input_file}")
print(f"출력: {output_file}")
if backup and Path(output_file).exists():
print(f"(파일이 존재했다면 백업이 생성되었습니다)")
print()
def main():
parser = argparse.ArgumentParser(
description='COCO Annotation Editing Utility',
formatter_class=argparse.RawDescriptionHelpFormatter,
epilog=__doc__
)
subparsers = parser.add_subparsers(dest='command', help='Available commands')
# info command
info_parser = subparsers.add_parser('info', help='Show annotation info')
info_parser.add_argument('-i', '--input', required=True, help='Input annotation file')
# remove command
remove_parser = subparsers.add_parser('remove', help='Remove categories')
remove_parser.add_argument('-i', '--input', required=True, help='Input annotation file')
remove_parser.add_argument('-o', '--output', required=True, help='Output annotation file')
remove_parser.add_argument('-c', '--classes', required=True, help='Classes to remove (comma-separated)')
remove_parser.add_argument('--no-backup', action='store_true', help='Do not create backup')
# merge command
merge_parser = subparsers.add_parser('merge', help='Merge categories')
merge_parser.add_argument('-i', '--input', required=True, help='Input annotation file')
merge_parser.add_argument('-o', '--output', required=True, help='Output annotation file')
merge_parser.add_argument('-s', '--source', required=True, help='Source classes to merge (comma-separated)')
merge_parser.add_argument('-t', '--target', required=True, help='Target class name after merge')
merge_parser.add_argument('--supercategory', default='', help='Supercategory for merged class')
merge_parser.add_argument('--no-backup', action='store_true', help='Do not create backup')
# rename command
rename_parser = subparsers.add_parser('rename', help='Rename categories')
rename_parser.add_argument('-i', '--input', required=True, help='Input annotation file')
rename_parser.add_argument('-o', '--output', required=True, help='Output annotation file')
rename_parser.add_argument('-m', '--mapping', required=True,
help='Name mapping (old:new,old2:new2 format)')
rename_parser.add_argument('--no-backup', action='store_true', help='Do not create backup')
# reindex command
reindex_parser = subparsers.add_parser('reindex', help='Reindex category IDs')
reindex_parser.add_argument('-i', '--input', required=True, help='Input annotation file')
reindex_parser.add_argument('-o', '--output', required=True, help='Output annotation file')
reindex_parser.add_argument('--start', type=int, default=1, help='Starting ID (default: 1)')
reindex_parser.add_argument('--no-backup', action='store_true', help='Do not create backup')
# If no arguments provided, enter interactive mode
if len(sys.argv) == 1:
interactive_mode()
return
args = parser.parse_args()
if not args.command:
parser.print_help()
return
# Execute command
if args.command == 'info':
editor = COCOAnnotationEditor(args.input)
editor.print_info()
elif args.command == 'remove':
classes = [c.strip() for c in args.classes.split(',')]
print(f"\n[*] 클래스 제거 중...")
editor = COCOAnnotationEditor(args.input)
editor.remove_categories(classes)
editor.save(args.output, backup=not args.no_backup)
elif args.command == 'merge':
source_classes = [c.strip() for c in args.source.split(',')]
print(f"\n[*] 클래스 통합 중...")
editor = COCOAnnotationEditor(args.input)
editor.merge_categories(source_classes, args.target, '')
editor.save(args.output, backup=not args.no_backup)
elif args.command == 'rename':
mapping = {}
for pair in args.mapping.split(','):
old, new = pair.split(':')
mapping[old.strip()] = new.strip()
print(f"\n[*] 클래스 이름 변경 중...")
editor = COCOAnnotationEditor(args.input)
editor.rename_categories(mapping)
editor.save(args.output, backup=not args.no_backup)
elif args.command == 'reindex':
print(f"\n[*] 클래스 ID 재할당 중...")
editor = COCOAnnotationEditor(args.input)
editor.reindex_categories(start_id=args.start)
editor.save(args.output, backup=not args.no_backup)
if __name__ == '__main__':
main()