小编War*_*ice的帖子

如何解决'CUDA内存不足。尝试在 pytorch 中分配 xxx MiB' ?

我正在尝试在 pytorch 中训练 CNN,但遇到了一些问题。运行时错误:

运行时错误:CUDA 内存不足。尝试分配 512.00 MiB(GPU 0;2.00 GiB 总容量;已分配 584.97 MiB;13.81 MiB 空闲;PyTorch 总共保留 590.00 MiB)

这是我的代码:

import os
import numpy as np
import cv2
import torch as t
import torch.nn as nn
import torchvision.transforms as transforms
from torch.utils.data import DataLoader,Dataset
import time
import matplotlib.pyplot as plt
%matplotlib inline
root_path='C:/Users/60960/Desktop/recet-task/course_LeeML20/course_LeeML20-datasets/hw3/food-11'
training_path=root_path+'/training'
testing_path=root_path+'/testing'
validation_path=root_path+'/validation'
def readfile(path,has_label):
    img_paths=sorted(os.listdir(path))
    x=np.zeros((len(img_paths),128,128,3),dtype=np.uint8)
    y=np.zeros((len(img_paths)),dtype=np.uint8)
    for i,file in enumerate(img_paths):
        img=cv2.imread(path+'/'+file)
        x[i,:,:]=cv2.resize(img,(128,128))
        if has_label:
            y[i]=int(file.split('_')[0])
    if has_label:
        return x,y
    else:
        return x
def show_img(img_from_cv2):
    b,g,r=cv2.split(img_from_cv2)
    img=cv2.merge([r,g,b]) …
Run Code Online (Sandbox Code Playgroud)

nvidia machine-learning deep-learning conv-neural-network pytorch

11
推荐指数
2
解决办法
6万
查看次数