這是一個(gè)工具包，功能：反向操作transforms.Normalize和transforms.ToTensor函數(shù)，將tensor格式的圖片轉(zhuǎn)換成.jpg、.png格式的圖片

?（注：這里是我原始的寫(xiě)法，但是是存在著一些改進(jìn)空間的【如評(píng)論區(qū)所言，會(huì)改變img_tensor的內(nèi)容，但我這邊就暫時(shí)保持原樣了，詳細(xì)的改進(jìn)方法可見(jiàn)評(píng)論區(qū)】【感謝評(píng)論區(qū)大佬！】）

import torch import numpy as np from PIL import Image import torchvision.transforms as transformsdef transform_convert(img_tensor, transform):"""param img_tensor: tensorparam transforms: torchvision.transforms"""if 'Normalize' in str(transform):normal_transform = list(filter(lambda x: isinstance(x, transforms.Normalize), transform.transforms))mean = torch.tensor(normal_transform[0].mean, dtype=img_tensor.dtype, device=img_tensor.device)std = torch.tensor(normal_transform[0].std, dtype=img_tensor.dtype, device=img_tensor.device)img_tensor.mul_(std[:,None,None]).add_(mean[:,None,None])img_tensor = img_tensor.transpose(0,2).transpose(0,1) # C x H x W ---> H x W x Cif 'ToTensor' in str(transform) or img_tensor.max() < 1:img_tensor = img_tensor.detach().numpy()*255if isinstance(img_tensor, torch.Tensor):img_tensor = img_tensor.numpy()if img_tensor.shape[2] == 3:img = Image.fromarray(img_tensor.astype('uint8')).convert('RGB')elif img_tensor.shape[2] == 1:img = Image.fromarray(img_tensor.astype('uint8')).squeeze()else:raise Exception("Invalid img shape, expected 1 or 3 in axis 2, but got {}!".format(img_tensor.shape[2]))return img

代碼示例

import matplotlib.pyplot as plt img = Image.open('./lena.jpg') ToTensor_transform = transforms.Compose([transforms.ToTensor()]) img_tensor = ToTensor_transform(img) img = transform_convert(img_tensor, ToTensor_transform) plt.imshow(img) plt.savefig('./lenaa.png')

總結(jié)

以上是生活随笔為你收集整理的将tensor张量转换成图片格式并保存的全部?jī)?nèi)容，希望文章能夠幫你解決所遇到的問(wèn)題。

如果覺(jué)得生活随笔網(wǎng)站內(nèi)容還不錯(cuò)，歡迎將生活随笔推薦給好友。