修改保存图片节点
This commit is contained in:
parent
f52db91713
commit
a06de2445c
|
|
@ -1,36 +1,33 @@
|
|||
import os.path
|
||||
import os
|
||||
import uuid
|
||||
|
||||
import torch
|
||||
import torchvision
|
||||
|
||||
|
||||
from PIL import Image
|
||||
import numpy as np
|
||||
class SaveImagePath:
|
||||
@classmethod
|
||||
def INPUT_TYPES(s):
|
||||
return {
|
||||
"required": {
|
||||
"image_path":("IMAGE", {"forceInput": True}),
|
||||
"image_path": ("IMAGE", {"forceInput": True}),
|
||||
}
|
||||
}
|
||||
|
||||
RETURN_TYPES = ("STRING",)
|
||||
|
||||
FUNCTION = "load"
|
||||
|
||||
CATEGORY = "不忘科技-自定义节点🚩"
|
||||
|
||||
def load(self, image_path:torch.Tensor):
|
||||
image_path = image_path.float()
|
||||
# 假设数据范围在 [0, 255],进行归一化
|
||||
image_path = image_path / 255.0
|
||||
# 检查并调整形状(这里只是示例,具体调整需根据实际情况)
|
||||
if len(image_path.shape) == 3:
|
||||
image_path = image_path.unsqueeze(0) # 添加批次维度
|
||||
u = uuid.uuid4()
|
||||
p = os.path.join(os.path.dirname(os.path.abspath(__file__)), "output", "%s.jpg" % str(u))
|
||||
torchvision.utils.save_image(image_path, p)
|
||||
# u = uuid.uuid4()
|
||||
# p = os.path.join(os.path.dirname(os.path.abspath(__file__)),"output","%s.jpg" % str(u))
|
||||
# torchvision.utils.save_image(image_path, p)
|
||||
return (p,)
|
||||
def load(self, image_path):
|
||||
# 确保数据类型为uint8且在0 - 255范围内
|
||||
if image_path.dtype != np.uint8:
|
||||
image_path = np.clip(image_path, 0, 255).astype(np.uint8)
|
||||
# 如果是单通道图像,转换为3通道
|
||||
if len(image_path.shape) == 2:
|
||||
image_path = np.stack([image_path] * 3, axis=-1)
|
||||
# 如果是通道优先格式 (C, H, W),转换为通道最后格式 (H, W, C)
|
||||
elif len(image_path.shape) == 3 and image_path.shape[0] <= 4:
|
||||
image_path = np.transpose(image_path, (1, 2, 0))
|
||||
pil_image = Image.fromarray(image_path)
|
||||
output_dir = os.path.join(os.path.dirname(os.path.abspath(__file__)), "output")
|
||||
if not os.path.exists(output_dir):
|
||||
os.makedirs(output_dir)
|
||||
file_name = "%s.jpg" % str(uuid.uuid4())
|
||||
p = os.path.join(output_dir, file_name)
|
||||
pil_image.save(p)
|
||||
return (p,)
|
||||
Loading…
Reference in New Issue