weixin_43745072的博客

使用DataParallel保存的模型在保存后载入的时候存在权重不匹配的问题，原因是每层的参数是会比原来多一个module的设置。import torchfrom collections import OrderedDict#python自带的库state_dict=torch.load(config.model_save_path+"_best")["model"]#我的权重存储在文件的model字段里面的'''state = { 'model': model

今天在进行torch模型的初始化的时候，发现报错：Object has no attribute ‘weight’回顾模型，发现在模型权重初始化函数，定义的带有conv的层的初始化是这样的。def weights_init(m): """init the weight for a network""" classname=m.__class__.__name__ # print(classname) if classname.find("conv")!=-1:

今天在读取png，在将png读入之后，想做一个/255并保存为.tif格式，出现大问题。写入图片import cv2 import numpy as npfrom google.colab.patches import cv2_imshowimg =cv2.imread("/content/drive/MyDrive/mathcup/B/t1/107.png",2)# 1彩色2灰度img=img/255cv2.imwrite("/content/t1.tif",img,[int(cv2

import torchimport torch.nn.functional as Fimport numpy as npa = torch.Tensor(128,1,512)B = a.repeat(1,5,1)print(B.shape[0])print(B.shape)128torch.Size([128, 5, 512])