如何从DataLoader获取样本的文件名? [英] How to get the filename of a sample from a DataLoader?

问题描述

我需要编写一个文件，其中包含我训练的卷积神经网络的数据测试结果.该数据包括语音数据收集.文件格式需要为文件名，预测"，但是我很难提取文件名.我这样加载数据:

I need to write a file with the result of the data test of a Convolutional Neural Network that I trained. The data include speech data collection. The file format needs to be "file name, prediction", but I am having a hard time to extract the file name. I load the data like this:

import torchvision
from torchvision import transforms
from torch.utils.data import DataLoader

TEST_DATA_PATH = ...

trans = transforms.Compose([
    transforms.ToTensor(),
    transforms.Normalize((0.1307,), (0.3081,))
])

test_dataset = torchvision.datasets.MNIST(
    root=TEST_DATA_PATH,
    train=False,
    transform=trans,
    download=True
)

test_loader = DataLoader(dataset=test_dataset, batch_size=1, shuffle=False)

并且我正在尝试按以下方式写入文件:

and I am trying to write to the file as follows:

f = open("test_y", "w")
with torch.no_grad():
    for i, (images, labels) in enumerate(test_loader, 0):
        outputs = model(images)
        _, predicted = torch.max(outputs.data, 1)
        file = os.listdir(TEST_DATA_PATH + "/all")[i]
        format = file + ", " + str(predicted.item()) + '\n'
        f.write(format)
f.close()

os.listdir(TESTH_DATA_PATH + "/all")[i]的问题在于它与test_loader的加载文件顺序不同步.我该怎么办?

The problem with os.listdir(TESTH_DATA_PATH + "/all")[i] is that it is not synchronized with the loaded files order of test_loader. What can I do?

推荐答案

通常情况下 DataLoader 可以从其中提供的数据集中为您提供批次.

In general case DataLoader is there to provide you the batches from the Dataset(s) it has inside.

在出现单标签/多标签分类问题时提到了AS @Barriel，DataLoader没有图像文件名，只有表示图像的张量以及类/标签.

AS @Barriel mentioned in case of single/multi-label classification problems, the DataLoader doesn't have image file name, just the tensors representing the images , and the classes / labels.

但是，DataLoader构造函数在加载对象时可能会占用很小的空间(与数据集一起，如果需要，您可以打包目标/标签和文件名)，甚至是数据框

However, DataLoader constructor when loading objects can take small things (together with the Dataset you may pack the targets/labels and the file names if you like) , even a dataframe

这样，DataLoader可能会以某种方式满足您的需求.

This way, the DataLoader may somehow grab that what you need.

这篇关于如何从DataLoader获取样本的文件名?的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！