labelme_to_coco(多个json文件转换成一个）
我是用labelme标注的动物数据集生成多个身体关键点json文件，然後我想转换成coco格式并合并成一個类似animal_keypoints_train.json文件（类似下面的配置信息）
{"image_id": 1448, "bbox": [234, 11, 344, 248], "keypoints": [[298, 45, 1], [274, 49, 1], [300, 59, 1], [294, 29, 1], [259, 32, 1], [293, 150, 1], [256, 156, 1], [330, 166, 1], [307, 173, 1], [296, 207, 1], [267, 206, 1], [325, 212, 1], [0, 0, 0], [292, 239, 1], [261, 239, 1], [320, 232, 1], [292, 234, 1], [274, 106, 1], [0, 0, 0], [0, 0, 0]], "num_keypoints": 20, "category_id": 1}, {"image_id": 1449, "bbox": [140, 26, 500, 333], "keypoints": [[257, 130, 1], [167, 113, 1], [177, 176, 1], [338, 113, 1], [173, 76, 1], [0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0], [262, 242, 1], [376, 168, 1], [0, 0, 0]], "num_keypoints": 20, "category_id": 1}, {"image_id": 1450, "bbox": [79, 1, 435, 334], "keypoints": [[336, 144, 1], [208, 124, 1], [298, 183, 1], [357, 101, 1], [147, 71, 1], [0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0], [249, 318, 1], [0, 0, 0], [0, 0, 0]], "num_keypoints": 20, "category_id": 2}, {"image_id": 1451, "bbox": [110, 158, 299, 267], "keypoints": [[0, 0, 0], [279, 210, 1], [286, 220, 1], [266, 169, 1], [266, 187, 1], [0, 0, 0], [235, 253, 1], [173, 198, 1], [116, 242, 1], [0, 0, 0], [273, 246, 1], [0, 0, 0], [0, 0, 0], [0, 0, 0], [291, 246, 1], [0, 0, 0], [0, 0, 0], [250, 226, 1], [213, 214, 1], [128, 195, 1]], "num_keypoints": 20, "category_id": 2},
{"supercategory": "animal", "id": 4, "name": "horse", "keypoints": ["left_eye", "right_eye", "nose", "left_ear", "right_ear", "left_front_elbow", "right_front_elbow", "left_back_elbow", "right_back_elbow", "left_front_knee", "right_front_knee", "left_back_knee", "right_back_knee", "left_front_paw", "right_front_paw", "left_back_paw", "right_back_paw", "throat", "withers", "tailbase"], "skeleton": [[0, 1], [0, 2], [1, 2], [0, 3], [1, 4], [2, 17], [18, 19], [5, 9], [6, 10], [7, 11], [8, 12], [9, 13], [10, 14], [11, 15], [12, 16]]},
结全ChatGPT和我自己的想法：
将从LabelMe标注生成的多个JSON文件转换为单个COCO格式的JSON文件，您可以按照以下步骤使用Python进行操作。我将为您提供一个脚本，该脚本会读取多个LabelMe JSON文件，提取相关信息，然后创建一个单独的COCO格式的JSON文件。
假设您有一个包含LabelMe JSON文件的目录，以及您提供的COCO格式JSON模式，您可以使用以下脚本：
import json
import os
# LabelMe JSON文件列表
labelme_json_files = [
'path/to/labelme_file1.json',
'path/to/labelme_file2.json',
# 根据需要添加更多文件路径
]
# COCO格式骨架和关键点信息
coco_keypoints_info = {
# 在此处填写您的COCO关键点信息
}
# 初始化COCO注释格式
coco_data = {
"images": [],
"annotations": [],
"categories": [coco_keypoints_info],
}
# 加载并处理每个LabelMe JSON文件
image_id_counter = 1
annotation_id_counter = 1
for labelme_file_path in labelme_json_files:
with open(labelme_file_path, 'r') as f:
labelme_data = json.load(f)
# 提取图像信息
image_info = {
"id": image_id_counter,
"width": labelme_data['imageWidth'],
"height": labelme_data['imageHeight'],
"file_name": os.path.basename(labelme_file_path).replace('.json', '.jpg'),
}
coco_data['images'].append(image_info)
# 处理注释
for annotation in labelme_data['shapes']:
# 提取注释信息
bbox = annotation['points']
keypoints = [] # 从LabelMe格式提取关键点
# 将关键点转换为COCO格式
# 创建COCO注释
coco_annotation = {
"id": annotation_id_counter,
"image_id": image_id_counter,
"category_id": 1, # 根据需要替换为适当的category_id
"bbox": [bbox[0][0], bbox[0][1], bbox[1][0] - bbox[0][0], bbox[1][1] - bbox[0][1]],
"keypoints": keypoints,
"num_keypoints": len(keypoints) // 3,
}
coco_data['annotations'].append(coco_annotation)
annotation_id_counter += 1
image_id_counter += 1
# 保存COCO格式JSON
output_json_path = 'path/to/animal_keypoints_train.json'
with open(output_json_path, 'w') as f:
json.dump(coco_data, f, indent=4)
请注意，您需要根据实际的目录结构、路径和数据格式调整此脚本。此外，请确保您根据LabelMe数据正确填充脚本中的keypoints列表。
【以下回答由 GPT 生成】
我可以帮助你将多个labelme标注的json文件转换为单个coco格式的文件。下面是解决该问题的Python代码：
import json
import os
from collections import defaultdict
# 读取所有json文件
json_files = ["json_file1.json", "json_file2.json", "json_file3.json"]
# 创建COCO格式的字典
coco_data = {
"images": [],
"annotations": [],
"categories": []
}
# 用于映射标签名和ID的字典
label_id_map = {
"animal": 1,
"human": 2,
# 添加其他标签
}
# 创建用于存储图像ID和标注ID的变量
image_id = 1
annotation_id = 1
# 遍历所有json文件
for json_file in json_files:
with open(json_file, "r") as f:
data = json.load(f)
# 添加类别信息到COCO格式字典的categories字段
category_id = len(coco_data["categories"]) + 1
category_name = data["label_name"]
category_info = {
"id": category_id,
"name": category_name,
"supercategory": "object"
}
coco_data["categories"].append(category_info)
# 遍历每个标注信息
for annotation in data["annotations"]:
# 创建图像信息
image_info = {
"id": image_id,
"file_name": data["image_name"],
"width": data["image_width"],
"height": data["image_height"]
}
coco_data["images"].append(image_info)
# 创建标注信息
annotation_info = {
"id": annotation_id,
"image_id": image_id,
"category_id": category_id,
"bbox": annotation["bbox"],
"keypoints": annotation["keypoints"],
"num_keypoints": annotation["num_keypoints"]
}
coco_data["annotations"].append(annotation_info)
# 增加图像ID和标注ID
image_id += 1
annotation_id += 1
# 将COCO格式字典保存为json文件
with open("animal_keypoints_train.json", "w") as f:
json.dump(coco_data, f)
请注意，上述代码假设labelme生成的每个json文件只包含一个图像的标注信息。如果有多个图像的标注信息存在于同一个json文件中，可以对代码进行相应的修改和调整。
希望这段代码能够帮助你解决问题！如果你有任何进一步的问题，请随时提问。
如果你已经解决了该问题, 非常希望你能够分享一下解决方案, 写成博客, 将相关链接放在评论区, 以帮助更多的人 ^-^
我还没有现成的工具可以转，我是通过自己写的py代码转的。
单个动物的JSON转换为COCO格式
labelme_json = {
"imagePath": "image.jpg",
"shapes": [], # Your shapes data here
}
coco_annotation = {
"image_id": 1448, # Replace with actual image_id
"bbox": [234, 11, 344, 248], # Replace with actual bbox
"keypoints": [], # Your keypoints data here
"num_keypoints": 20, # Replace with actual num_keypoints
"category_id": 1, # Replace with actual category_id
}
# Now combine labelme_json and coco_annotation to create COCO-like annotation
coco_annotation["image_id"] = labelme_json["imagePath"] # Use image path as image_id
coco_annotation["keypoints"] = [] # Convert keypoints data from labelme_json to COCO format
# Add more fields as needed
# Now coco_annotation is your COCO-like annotation for one animal
合并成一个文件
import json
combined_data = [] # List to store combined data
# Load and combine each JSON file
for filename in ['animal1.json', 'animal2.json', ...]: # Replace with your filenames
with open(filename, 'r') as json_file:
data = json.load(json_file)
combined_data.extend(data) # Append data to the combined list
# Save the combined data as a new JSON file
with open('animal_keypoints_train.json', 'w') as output_file:
json.dump(combined_data, output_file)
你可以使用Python来将多个LabelMe标注的JSON文件转换成COCO格式，并将它们合并成一个类似animal_keypoints_train.json的文件。下面是一个示例代码，你需要根据你的实际数据结构进行适当的调整：
import json
# 初始化COCO格式的数据
coco_data = {
"images": [],
"annotations": [],
"categories": []
}
# 定义类别信息
category_info = {
"supercategory": "animal",
"id": 1,
"name": "animal",
"keypoints": ["left_eye", "right_eye", "nose", ...], # 你的关键点信息
"skeleton": [[0, 1], [0, 2], [1, 2], ...] # 你的关键点连接信息
}
coco_data["categories"].append(category_info)
# 读取多个LabelMe JSON文件并转换为COCO格式
json_filenames = ["file1.json", "file2.json", ...] # 替换为你的文件名
image_id = 1 # 初始化image_id
for json_filename in json_filenames:
with open(json_filename, "r") as json_file:
labelme_data = json.load(json_file)
# 处理每个图像
for labelme_image in labelme_data["images"]:
image_info = {
"id": image_id,
"file_name": labelme_image["file_name"],
"height": labelme_image["height"],
"width": labelme_image["width"]
}
coco_data["images"].append(image_info)
# 处理每个关键点标注
for annotation in labelme_data["annotations"]:
# 假设你的关键点数据在"keypoints"字段中
keypoints = annotation["keypoints"]
num_keypoints = keypoints.count(2) # 统计关键点个数
# 构建COCO格式的关键点信息
coco_annotation = {
"image_id": image_id,
"bbox": annotation["bbox"],
"keypoints": keypoints,
"num_keypoints": num_keypoints,
"category_id": 1 # 假设都是同一个类别
}
coco_data["annotations"].append(coco_annotation)
image_id += 1
# 将合并后的COCO数据写入文件
with open("animal_keypoints_train.json", "w") as output_file:
json.dump(coco_data, output_file)
问题点:多个身体关键点json文件合并成一個animal_keypoints_train.json文件
代码如下
import json
def save_json(coco_data):
# 将COCO格式字典保存为json文件
with open("animal_keypoints_train.json", "w") as f:
json.dump(coco_data, f)
# 合并各个身体关键点json文件
def update_data(file_path, image_id, annotation_id, coco_data):
with open(file_path, "r") as f:
data = json.load(f)
# 添加类别信息到COCO格式字典的categories字段
category_item_id = len(coco_data["categories"]) + 1
name = data["label_name"]
category_info = {
"id": category_item_id,
"name": name,
"supercategory": "none"
}
coco_data["categories"].append(category_info)
# 遍历每个标注信息
for annotation in data["annotations"]:
# 创建图像信息
image_info = {
"id": image_id,
"file_name": data["image_name"],
"width": data["image_width"],
"height": data["image_height"]
}
coco_data["images"].append(image_info)
# 创建标注信息
annotation_info = {
"id": annotation_id,
"image_id": image_id,
"category_id": category_item_id,
"bbox": annotation["bbox"],
"keypoints": annotation["keypoints"],
"num_keypoints": annotation["num_keypoints"]
}
coco_data["annotations"].append(annotation_info)
def run():
# 创建COCO格式的字典
coco_data = {
"images": [],
"type": "instances",
"annotations": [],
"categories": []
}
# 读取所有json文件
files = ["file1.json", "file2.json"]
for file_id, file_path in enumerate(files, 1):
# 文件处理顺序 作为图像ID和标注ID的变量
update_data(file_path, file_id, file_id, coco_data)
# 存成json文件
save_json(coco_data)
参考gpt：
结合自己分析给你如下建议：
labelme是一种常用的图像标注工具，它可以生成json格式的标注文件，包含多边形、矩形、圆形、线条和点等形状。
coco是一种常用的图像数据集格式，它可以用于目标检测、实例分割、关键点检测等任务。coco格式的数据集包含images、categories和annotations三个主要字段，其中annotations字段包含bbox、keypoints、num_keypoints和category_id等信息。
要将labelme标注的数据转换为coco格式的数据集，可以使用一些现有的工具或代码，例如labelme2coco4。这些工具或代码可以根据labelme生成的json文件，提取出图像的路径、大小、类别和关键点等信息，并将其转换为coco格式的json文件。
labelme标注后多个json文件转标准coco格式
类似得，可以参考下
https://codeleading.com/article/56216148954/#google_vignette
引用 皆我百晓生 小程序回复内容作答：
您可以使用以下Python代码将多个json文件转换为一个COCO格式的json文件：
import json
import os
def labelme_to_coco(json_files, output_file):
coco_data = {
"info": {"description": "Animal Keypoints Dataset", "version": "1.0", "year": 2021, "contributor": "Anonymous", "date_created": "2021/01/01"},
"licenses": [{"id": 1, "name": "Unknown License", "url": ""}],
"categories": [],
"images": [],
"annotations": []
}
category_id = 1
for json_file in json_files:
with open(json_file, "r") as f:
data = json.load(f)
# Add categories
coco_data["categories"].append({"id": category_id, "name": data["name"], "keypoints": data["keypoints"], "skeleton": data["skeleton"]})
# Add images and annotations
for image_data in data["images"]:
image_id = len(coco_data["images"]) + 1
image_info = {"id": image_id, "file_name": image_data["file_name"], "width": image_data["width"], "height": image_data["height"], "license": 1}
coco_data["images"].append(image_info)
for annotation in image_data["annotations"]:
bbox = annotation["bbox"]
keypoints = annotation["keypoints"]
num_keypoints = annotation["num_keypoints"]
coco_annotation = {"image_id": image_id, "bbox": bbox, "keypoints": keypoints, "num_keypoints": num_keypoints, "category_id": category_id}
coco_data["annotations"].append(coco_annotation)
category_id += 1
with open(output_file,"w") as f:
json.dump(coco_data, f)
# Example usage
json_files = ["file1.json", "file2.json", "file3.json"]
output_file = "animal_keypoints_train.json"
labelme_to_coco(json_files, output_file)
请确保将json_files列表替换为实际的文件名列表，并将output_file替换为输出的COCO格式json文件的文件名。
这段代码将遍历所有的json文件，将每个文件中的信息提取出来，并按照COCO格式构建新的json数据。然后将生成的json数据保存到输出文件中。