503 lines
22 KiB
Python
503 lines
22 KiB
Python
import json
|
||
import re
|
||
import math
|
||
from typing import List, Tuple
|
||
|
||
class StringToBbox:
|
||
@classmethod
|
||
def INPUT_TYPES(s):
|
||
return {
|
||
"required": {
|
||
"string": ("STRING", {
|
||
"multiline": True,
|
||
#"default": '[\n {"bbox_2d": [284, 252, 488, 815], "label": "凌波丽"}\n]',
|
||
#"placeholder": "输入JSON格式的bbox字符串"
|
||
}),
|
||
"width": ("INT", {
|
||
"default": 1024,
|
||
"min": 1,
|
||
"max": 10000,
|
||
"step": 1,
|
||
#"description": "图片的实际宽度（像素）"
|
||
}),
|
||
"height": ("INT", {
|
||
"default": 1024,
|
||
"min": 1,
|
||
"max": 10000,
|
||
"step": 1,
|
||
#"description": "图片的实际高度（像素）"
|
||
}),
|
||
"index": ("STRING", {
|
||
"default": "",
|
||
#"placeholder": "指定要提取的bbox索引，多个用逗号分隔（如0,1,2）"
|
||
}),
|
||
},
|
||
}
|
||
|
||
RETURN_TYPES = ("STRING", "BBOX")
|
||
RETURN_NAMES = ("center_coordinates", "bboxes")
|
||
FUNCTION = "convert_to_bbox"
|
||
CATEGORY = "image/caption"
|
||
#DESCRIPTION = "将JSON格式的千分比bbox字符串转换为实际像素坐标的BBOX"
|
||
|
||
def _clean_and_extract_coords(self, input_string: str) -> List[List[float]]:
|
||
all_bboxes = []
|
||
|
||
# 1. 使用更稳健的正则：直接从原始字符串中提取所有非嵌套的 [ ... ] 内容。
|
||
# 正则 r'\[([^\[\]]+)\]' 匹配 [ 后面跟着任何非 [ 或 ] 的字符，直到下一个 ]。
|
||
# 这样可以忽略包含 JSON 键值对的外层结构，专门提取内层的数字列表。
|
||
# 这样可以同时处理：
|
||
# - 纯数组输入：[[250, 150, 450, 500], [600, 700, 800, 950]]
|
||
# - JSON 对象输入：{"bbox_2d": [284, 252, 488, 815], ...}
|
||
bracket_groups = re.findall(r'\[([^\[\]]+)\]', input_string, re.DOTALL) #
|
||
|
||
# 2. 解析每个数组，筛选有效bbox（长度为4的数值数组）
|
||
for group in bracket_groups:
|
||
# 针对捕获的内容进行局部清理，只保留数字、逗号、小数点、负号
|
||
# 这样可以在不破坏整体结构的情况下，清理掉例如 "label" 或其他文本
|
||
cleaned_group = re.sub(r'[^-0-9,.]', '', group)
|
||
|
||
# 分割数值字符串，过滤空字符串
|
||
num_strings = [s.strip() for s in cleaned_group.split(',') if s.strip()]
|
||
nums = []
|
||
for num_str in num_strings:
|
||
# 验证是否为有效数字（整数/小数）
|
||
if re.match(r'^-?\d+(\.\d+)?$', num_str):
|
||
nums.append(float(num_str))
|
||
|
||
# 只保留长度为4的数组（符合bbox坐标格式）
|
||
if len(nums) == 4:
|
||
# 去重
|
||
if nums not in all_bboxes:
|
||
all_bboxes.append(nums)
|
||
|
||
# 调试输出仅用于检查最终结果
|
||
# print(f"清理后的字符串：{input_string.strip()}") # 保持原始字符串
|
||
|
||
return all_bboxes
|
||
|
||
def convert_to_bbox(self, string: str, width: int, height: int, index: str) -> Tuple[str, List[List[int]]]:
|
||
"""
|
||
核心转换函数：清理字符串→提取坐标→按规则转换为像素坐标
|
||
"""
|
||
# 初始化返回值
|
||
center_points = []
|
||
bboxes = []
|
||
|
||
try:
|
||
# 1. 清理字符串并提取所有有效的bbox坐标数组
|
||
all_bboxes = self._clean_and_extract_coords(string)
|
||
if not all_bboxes:
|
||
raise ValueError("未从输入字符串中提取到有效的bbox坐标（需包含长度为4的数值数组）")
|
||
|
||
# 2. 处理索引参数
|
||
if index.strip():
|
||
# 解析用户指定的索引（支持多个，用逗号分隔）
|
||
indexes = []
|
||
for idx_str in index.split(","):
|
||
idx_str = idx_str.strip()
|
||
if idx_str.isdigit():
|
||
indexes.append(int(idx_str))
|
||
else:
|
||
raise ValueError(f"无效的索引值：{idx_str}（必须是整数）")
|
||
else:
|
||
# 如果索引为空，使用所有可用的bbox索引
|
||
indexes = list(range(len(all_bboxes)))
|
||
|
||
# 3. 遍历指定的索引，转换坐标
|
||
for idx in indexes:
|
||
if 0 <= idx < len(all_bboxes):
|
||
# 获取千分比坐标（左上x, 左上y, 右下x, 右下y）
|
||
x1_percent, y1_percent, x2_percent, y2_percent = all_bboxes[idx]
|
||
|
||
# 转换为实际像素坐标（四舍五入为整数）
|
||
x1 = int(math.floor(x1_percent * width / 1000))
|
||
y1 = int(math.floor(y1_percent * height / 1000))
|
||
x2 = int(math.ceil(x2_percent * width / 1000))
|
||
y2 = int(math.ceil(y2_percent * height / 1000))
|
||
|
||
# 确保坐标在合理范围内（不小于0，不大于图片尺寸）
|
||
x1 = max(0, min(x1, width))
|
||
y1 = max(0, min(y1, height))
|
||
x2 = max(x1, min(x2, width)) # 确保x2 >= x1
|
||
y2 = max(y1, min(y2, height)) # 确保y2 >= y1
|
||
|
||
# 计算中心点坐标
|
||
center_x = int(round((x1 + x2) / 2))
|
||
center_y = int(round((y1 + y2) / 2))
|
||
|
||
# 添加到结果列表
|
||
center_points.append({"x": center_x, "y": center_y})
|
||
bboxes.append([x1, y1, x2, y2])
|
||
else:
|
||
raise ValueError(f"索引{idx}超出范围（有效范围：0-{len(all_bboxes)-1}）")
|
||
|
||
except json.JSONDecodeError as e:
|
||
# JSON解析错误
|
||
print(f"JSON解析错误：{str(e)}")
|
||
center_points = [{"x": 0, "y": 0}]
|
||
bboxes = [[0, 0, 0, 0]]
|
||
except ValueError as e:
|
||
# 自定义错误
|
||
print(f"参数错误：{str(e)}")
|
||
center_points = [{"x": 0, "y": 0}]
|
||
bboxes = [[0, 0, 0, 0]]
|
||
except Exception as e:
|
||
# 其他未知错误
|
||
print(f"转换错误：{str(e)}")
|
||
center_points = [{"x": 0, "y": 0}]
|
||
bboxes = [[0, 0, 0, 0]]
|
||
|
||
# 输出调试信息
|
||
#print(f"提取到的原始坐标（千分比）：{all_bboxes}")
|
||
#print(f"转换后的BBOXes（像素）：{bboxes}")
|
||
#print(f"中心点坐标：{center_points}")
|
||
|
||
# 返回结果（中心点JSON字符串 + BBOX列表）
|
||
return (json.dumps(center_points), bboxes)
|
||
|
||
|
||
class StringToComfyBbox:
|
||
@classmethod
|
||
def INPUT_TYPES(s):
|
||
return {
|
||
"required": {
|
||
"string": ("STRING", {
|
||
"multiline": True,
|
||
#"default": '[\n {"bbox_2d": [284, 252, 488, 815], "label": "凌波丽"}\n]',
|
||
#"placeholder": "输入JSON格式的bbox字符串"
|
||
}),
|
||
"width": ("INT", {
|
||
"default": 1024,
|
||
"min": 1,
|
||
"max": 10000,
|
||
"step": 1,
|
||
#"description": "图片的实际宽度（像素）"
|
||
}),
|
||
"height": ("INT", {
|
||
"default": 1024,
|
||
"min": 1,
|
||
"max": 10000,
|
||
"step": 1,
|
||
#"description": "图片的实际高度（像素）"
|
||
}),
|
||
"index": ("STRING", {
|
||
"default": "",
|
||
#"placeholder": "指定要提取的bbox索引，多个用逗号分隔（如0,1,2）"
|
||
}),
|
||
},
|
||
}
|
||
|
||
RETURN_TYPES = ("STRING", "BOUNDING_BOX")
|
||
RETURN_NAMES = ("center_coordinates", "bboxes")
|
||
FUNCTION = "convert_to_comfy_bbox"
|
||
CATEGORY = "image/caption"
|
||
#DESCRIPTION = "将JSON格式的千分比bbox字符串转换为实际像素坐标的BBOX"
|
||
|
||
def _clean_and_extract_coords(self, input_string: str) -> List[List[float]]:
|
||
all_bboxes = []
|
||
|
||
# 1. 使用更稳健的正则：直接从原始字符串中提取所有非嵌套的 [ ... ] 内容。
|
||
# 正则 r'\[([^\[\]]+)\]' 匹配 [ 后面跟着任何非 [ 或 ] 的字符，直到下一个 ]。
|
||
# 这样可以忽略包含 JSON 键值对的外层结构，专门提取内层的数字列表。
|
||
# 这样可以同时处理：
|
||
# - 纯数组输入：[[250, 150, 450, 500], [600, 700, 800, 950]]
|
||
# - JSON 对象输入：{"bbox_2d": [284, 252, 488, 815], ...}
|
||
bracket_groups = re.findall(r'\[([^\[\]]+)\]', input_string, re.DOTALL) #
|
||
|
||
# 2. 解析每个数组，筛选有效bbox（长度为4的数值数组）
|
||
for group in bracket_groups:
|
||
# 针对捕获的内容进行局部清理，只保留数字、逗号、小数点、负号
|
||
# 这样可以在不破坏整体结构的情况下，清理掉例如 "label" 或其他文本
|
||
cleaned_group = re.sub(r'[^-0-9,.]', '', group)
|
||
|
||
# 分割数值字符串，过滤空字符串
|
||
num_strings = [s.strip() for s in cleaned_group.split(',') if s.strip()]
|
||
nums = []
|
||
for num_str in num_strings:
|
||
# 验证是否为有效数字（整数/小数）
|
||
if re.match(r'^-?\d+(\.\d+)?$', num_str):
|
||
nums.append(float(num_str))
|
||
|
||
# 只保留长度为4的数组（符合bbox坐标格式）
|
||
if len(nums) == 4:
|
||
# 去重
|
||
if nums not in all_bboxes:
|
||
all_bboxes.append(nums)
|
||
|
||
# 调试输出仅用于检查最终结果
|
||
# print(f"清理后的字符串：{input_string.strip()}") # 保持原始字符串
|
||
|
||
return all_bboxes
|
||
|
||
def convert_to_comfy_bbox(self, string: str, width: int, height: int, index: str) -> Tuple[str, List[List[int]]]:
|
||
"""
|
||
核心转换函数：清理字符串→提取坐标→按规则转换为像素坐标
|
||
"""
|
||
# 初始化返回值
|
||
center_points = []
|
||
bboxes = []
|
||
|
||
try:
|
||
# 1. 清理字符串并提取所有有效的bbox坐标数组
|
||
all_bboxes = self._clean_and_extract_coords(string)
|
||
if not all_bboxes:
|
||
raise ValueError("未从输入字符串中提取到有效的bbox坐标（需包含长度为4的数值数组）")
|
||
|
||
# 2. 处理索引参数
|
||
if index.strip():
|
||
# 解析用户指定的索引（支持多个，用逗号分隔）
|
||
indexes = []
|
||
for idx_str in index.split(","):
|
||
idx_str = idx_str.strip()
|
||
if idx_str.isdigit():
|
||
indexes.append(int(idx_str))
|
||
else:
|
||
raise ValueError(f"无效的索引值：{idx_str}（必须是整数）")
|
||
else:
|
||
# 如果索引为空，使用所有可用的bbox索引
|
||
indexes = list(range(len(all_bboxes)))
|
||
|
||
# 3. 遍历指定的索引，转换坐标
|
||
for idx in indexes:
|
||
if 0 <= idx < len(all_bboxes):
|
||
# 获取千分比坐标（左上x, 左上y, 右下x, 右下y）
|
||
x1_percent, y1_percent, x2_percent, y2_percent = all_bboxes[idx]
|
||
|
||
# 转换为实际像素坐标（四舍五入为整数）
|
||
x1 = int(math.floor(x1_percent * width / 1000))
|
||
y1 = int(math.floor(y1_percent * height / 1000))
|
||
x2 = int(math.ceil(x2_percent * width / 1000))
|
||
y2 = int(math.ceil(y2_percent * height / 1000))
|
||
|
||
# 确保坐标在合理范围内（不小于0，不大于图片尺寸）
|
||
x1 = max(0, min(x1, width))
|
||
y1 = max(0, min(y1, height))
|
||
x2 = max(x1, min(x2, width)) # 确保x2 >= x1
|
||
y2 = max(y1, min(y2, height)) # 确保y2 >= y1
|
||
|
||
# 计算中心点坐标
|
||
center_x = int(round((x1 + x2) / 2))
|
||
center_y = int(round((y1 + y2) / 2))
|
||
|
||
# 添加到结果列表
|
||
center_points.append({"x": center_x, "y": center_y})
|
||
bboxes.append({"x": x1, "y": y1, "width": max(x2-x1,1), "height": max(y2-y1,1)})
|
||
else:
|
||
raise ValueError(f"索引{idx}超出范围（有效范围：0-{len(all_bboxes)-1}）")
|
||
|
||
except json.JSONDecodeError as e:
|
||
# JSON解析错误
|
||
print(f"JSON解析错误：{str(e)}")
|
||
center_points = [{"x": 0, "y": 0}]
|
||
bboxes = [{"x": 0, "y": 0, "width": 0, "height": 0}]
|
||
except ValueError as e:
|
||
# 自定义错误
|
||
print(f"参数错误：{str(e)}")
|
||
center_points = [{"x": 0, "y": 0}]
|
||
bboxes = [{"x": 0, "y": 0, "width": 0, "height": 0}]
|
||
except Exception as e:
|
||
# 其他未知错误
|
||
print(f"转换错误：{str(e)}")
|
||
center_points = [{"x": 0, "y": 0}]
|
||
bboxes = [{"x": 0, "y": 0, "width": 0, "height": 0}]
|
||
|
||
# 输出调试信息
|
||
#print(f"提取到的原始坐标（千分比）：{all_bboxes}")
|
||
#print(f"转换后的BBOXes（像素）：{bboxes}")
|
||
#print(f"中心点坐标：{center_points}")
|
||
|
||
# 返回结果（中心点JSON字符串 + BBOX列表）
|
||
return (json.dumps(center_points), bboxes)
|
||
|
||
|
||
class StringToSam3Box:
|
||
@classmethod
|
||
def INPUT_TYPES(s):
|
||
return {
|
||
"required": {
|
||
"string": ("STRING", {
|
||
"multiline": True,
|
||
}),
|
||
#"bboxes": ("BBOX",), # 接收 StringToBbox 的输出
|
||
# "width": ("INT", {
|
||
# "default": 1024,
|
||
# "min": 1,
|
||
# "max": 10000,
|
||
# "step": 1,
|
||
# "description": "原图宽度，用于归一化"
|
||
# }),
|
||
# "height": ("INT", {
|
||
# "default": 1024,
|
||
# "min": 1,
|
||
# "max": 10000,
|
||
# "step": 1,
|
||
# "description": "原图高度，用于归一化"
|
||
# }),
|
||
"label_type": (["positive", "negative"], {
|
||
"default": "positive"
|
||
}),
|
||
"index": ("STRING", {
|
||
"default": "",
|
||
})
|
||
},
|
||
}
|
||
|
||
RETURN_TYPES = ("SAM3_POINTS_PROMPT", "SAM3_BOXES_PROMPT",)
|
||
RETURN_NAMES = ("point_prompt", "box_prompt",)
|
||
FUNCTION = "convert_to_sam3box"
|
||
CATEGORY = "image/caption"
|
||
|
||
def _clean_and_extract_coords(self, input_string: str) -> List[List[float]]:
|
||
all_bboxes = []
|
||
|
||
# 1. 使用更稳健的正则：直接从原始字符串中提取所有非嵌套的 [ ... ] 内容。
|
||
# 正则 r'\[([^\[\]]+)\]' 匹配 [ 后面跟着任何非 [ 或 ] 的字符，直到下一个 ]。
|
||
# 这样可以忽略包含 JSON 键值对的外层结构，专门提取内层的数字列表。
|
||
# 这样可以同时处理：
|
||
# - 纯数组输入：[[250, 150, 450, 500], [600, 700, 800, 950]]
|
||
# - JSON 对象输入：{"bbox_2d": [284, 252, 488, 815], ...}
|
||
bracket_groups = re.findall(r'\[([^\[\]]+)\]', input_string, re.DOTALL) #
|
||
|
||
# 2. 解析每个数组，筛选有效bbox（长度为4的数值数组）
|
||
for group in bracket_groups:
|
||
# 针对捕获的内容进行局部清理，只保留数字、逗号、小数点、负号
|
||
# 这样可以在不破坏整体结构的情况下，清理掉例如 "label" 或其他文本
|
||
cleaned_group = re.sub(r'[^-0-9,.]', '', group)
|
||
|
||
# 分割数值字符串，过滤空字符串
|
||
num_strings = [s.strip() for s in cleaned_group.split(',') if s.strip()]
|
||
nums = []
|
||
for num_str in num_strings:
|
||
# 验证是否为有效数字（整数/小数）
|
||
if re.match(r'^-?\d+(\.\d+)?$', num_str):
|
||
nums.append(float(num_str))
|
||
|
||
# 只保留长度为4的数组（符合bbox坐标格式）
|
||
if len(nums) == 4:
|
||
# 去重
|
||
if nums not in all_bboxes:
|
||
all_bboxes.append(nums)
|
||
|
||
# 调试输出仅用于检查最终结果
|
||
# print(f"清理后的字符串：{input_string.strip()}") # 保持原始字符串
|
||
|
||
return all_bboxes
|
||
|
||
def convert_to_sam3box(self, string, label_type, index):
|
||
sam_labels = []
|
||
# 确定标签的布尔值
|
||
is_positive = False if label_type == "negative" else True
|
||
|
||
sam_boxes = []
|
||
center_points = []
|
||
|
||
try:
|
||
# 1. 清理字符串并提取所有有效的bbox坐标数组
|
||
all_bboxes = self._clean_and_extract_coords(string)
|
||
if not all_bboxes:
|
||
raise ValueError("未从输入字符串中提取到有效的bbox坐标（需包含长度为4的数值数组）")
|
||
|
||
# 2. 处理索引参数
|
||
if index.strip():
|
||
# 解析用户指定的索引（支持多个，用逗号分隔）
|
||
indexes = []
|
||
for idx_str in index.split(","):
|
||
idx_str = idx_str.strip()
|
||
if idx_str.isdigit():
|
||
indexes.append(int(idx_str))
|
||
else:
|
||
raise ValueError(f"无效的索引值：{idx_str}（必须是整数）")
|
||
else:
|
||
# 如果索引为空，使用所有可用的bbox索引
|
||
indexes = list(range(len(all_bboxes)))
|
||
|
||
# 3. 遍历指定的索引，转换坐标
|
||
for idx in indexes:
|
||
if 0 <= idx < len(all_bboxes):
|
||
# 获取千分比坐标（左上x, 左上y, 右下x, 右下y）
|
||
x1_percent, y1_percent, x2_percent, y2_percent = all_bboxes[idx]
|
||
|
||
# 转换为百分比像素坐标（小数）
|
||
if x2_percent > 1 or y2_percent > 1:
|
||
x1 = x1_percent / 1000
|
||
y1 = y1_percent / 1000
|
||
x2 = x2_percent / 1000
|
||
y2 = y2_percent / 1000
|
||
|
||
# 确保坐标在合理范围内（不小于0，不大于1）
|
||
x1 = max(0, min(x1, 1))
|
||
y1 = max(0, min(y1, 1))
|
||
x2 = max(0, min(x2, 1))
|
||
y2 = max(0, min(y2, 1))
|
||
|
||
# 计算中心点坐标
|
||
center_x = (x1 + x2) / 2
|
||
center_y = (y1 + y2) / 2
|
||
box_w = abs(x2 - x1)
|
||
box_h = abs(y2 - y1)
|
||
|
||
# 添加到结果列表
|
||
#center_points.append({"x": center_x, "y": center_y})
|
||
center_points.append([center_x, center_y])
|
||
sam_boxes.append([center_x, center_y, box_w, box_h])
|
||
sam_labels.append(is_positive)
|
||
else:
|
||
raise ValueError(f"索引{idx}超出范围（有效范围：0-{len(all_bboxes)-1}）")
|
||
|
||
except json.JSONDecodeError as e:
|
||
# JSON解析错误
|
||
print(f"JSON解析错误：{str(e)}")
|
||
center_points = [{"x": 0, "y": 0}]
|
||
bboxes = [[0, 0, 0, 0]]
|
||
except ValueError as e:
|
||
# 自定义错误
|
||
print(f"参数错误：{str(e)}")
|
||
center_points = [{"x": 0, "y": 0}]
|
||
bboxes = [[0, 0, 0, 0]]
|
||
except Exception as e:
|
||
# 其他未知错误
|
||
print(f"转换错误：{str(e)}")
|
||
center_points = [{"x": 0, "y": 0}]
|
||
bboxes = [[0, 0, 0, 0]]
|
||
|
||
# for bbox in bboxes:
|
||
# # bbox 格式通常为 [x1, y1, x2, y2] (像素值)
|
||
# x1, y1, x2, y2 = bbox
|
||
|
||
# # 计算宽度和高度 (像素)
|
||
# box_w = abs(x2 - x1)
|
||
# box_h = abs(y2 - y1)
|
||
|
||
# # 计算中心点 (像素)
|
||
# center_x = (x1 + x2) / 2
|
||
# center_y = (y1 + y2) / 2
|
||
|
||
# # 归一化 (转换为 0.0 - 1.0)
|
||
# # 格式: [center_x, center_y, width, height]
|
||
# norm_cx = center_x / width
|
||
# norm_cy = center_y / height
|
||
# norm_w = box_w / width
|
||
# norm_h = box_h / height
|
||
|
||
# # 确保数值在 0-1 之间 (防止因为浮点数计算超出极小范围)
|
||
# norm_cx = max(0.0, min(1.0, norm_cx))
|
||
# norm_cy = max(0.0, min(1.0, norm_cy))
|
||
# norm_w = max(0.0, min(1.0, norm_w))
|
||
# norm_h = max(0.0, min(1.0, norm_h))
|
||
|
||
#sam_boxes.append([norm_cx, norm_cy, norm_w, norm_h])
|
||
|
||
|
||
# 构建最终字典
|
||
result_boxes = {
|
||
"boxes": sam_boxes,
|
||
"labels": sam_labels
|
||
}
|
||
result_points = {
|
||
"points": center_points,
|
||
"labels": sam_labels
|
||
}
|
||
|
||
#return (json.dumps(result, indent=4),)
|
||
return (result_points, result_boxes) |