TypeError: __init__() 为参数 \'dim\' 获取了多个值

Question

我正在对两个训练有素的模型进行测试。首先，我在测试期间遇到错误，所以我将 torch.logsoftmax 类更改为 nn.LogSoftmax。

代码

from torch.utils.data import Dataset, DataLoader
import pandas as pd
from torchvision import transforms
from PIL import Image
import torch
import torch.nn as nn
from glob import glob
from pathlib import PurePath
import numpy as np
import timm
import torchvision
import time

img_list = glob('/media/cvpr/CM_22/OOD-CV-phase2/phase2-cls/images/*.jpg')

name_list = [
    'aeroplane',
    'bicycle',
    'boat',
    'bus',
    'car',
    'chair',
    'diningtable',
    'motorbike',
    'sofa',
    'train'
]

# conda install pytorch==1.9.0 torchvision==0.10.0 torchaudio==0.9.0 cudatoolkit=10.2 -c pytorch

class PoseData(Dataset):
    def __init__(self, transforms) -> None:
        """
        the data folder should look like
        - datafolder
            - Images
            - labels.csv        
        """
        super().__init__()
        self.img_list = glob('/media/cvpr/CM_22/OOD-CV-phase2/phase2-cls/images/*.jpg')
        self.img_list = sorted(self.img_list, key=lambda x: eval(PurePath(x).parts[-1][:-4]))
        self.trs = transforms

    def __len__(self):
        return len(self.img_list)

    def __getitem__(self, index):
        image_dir = self.img_list[index]
        image_name = PurePath(image_dir).parts[-1]
        image = Image.open(image_dir)
        image = self.trs(image)

        return image, image_name


if __name__ == "__main__":
    normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],
                                     std=[0.229, 0.224, 0.225])
    tfs = transforms.Compose([
        transforms.Resize((224, 224)),
        transforms.ToTensor(),
        normalize,
    ])

    model1 = timm.models.swin_base_patch4_window7_224(pretrained=False, num_classes=15)
    model1 = torch.nn.DataParallel(model1)
    model1.load_state_dict(torch.load('/media/cvpr/CM_22/OOD_CV/swin15_best.pth.tar')['state_dict'],strict=False)
    model1 = model1.cuda()
    model1.eval()

    model2 = timm.models.convnext_base(pretrained=False, num_classes=15)
    model2 = torch.nn.DataParallel(model2)
    model2.load_state_dict(torch.load('convnext15_best.pth.tar')['state_dict'],strict=False)
    model2 = model2.cuda()
    model2.eval()

    dataset = PoseData(tfs)
    loader = DataLoader(dataset, batch_size=128, shuffle=False, drop_last=False, num_workers=4)

    image_dir = []
    preds = []
    for image, pth in loader:
        image_dir.append(list(pth))
        image = image.cuda()

        with torch.no_grad():

            model1.eval()
            pred1 = model1(image)
            model2.eval()
            pred2 = model2(image)

            entropy1 = -torch.sum(torch.softmax(pred1[:, :10], dim=1) * nn.LogSoftmax(pred1[:, :10], dim=1), dim=-1,
                                  keep_dim=True)
            entropy2 = -torch.sum(torch.softmax(pred2[:, :10], dim=1) * nn.LogSoftmax(pred2[:, :10], dim=1), dim=-1,
                                  keep_dim=True)
            entropy = entropy1 + entropy2

            pred = torch.softmax(pred1[:, :10], dim=1) * (entropy - entropy1) / entropy + torch.softmax(pred2[:, :10],
                                                                                                        dim=1) * (
                               entropy - entropy2) / entropy
            pred = torch.argmax(pred[:, :10], dim=1)
            p = []
            for i in range(pred.size(0)):
                p.append(name_list[pred[i].item()])
        p = np.array(p)
        preds.append(p)
        print(len(np.concatenate(preds)))

    image_dir = np.array(sum(image_dir, []))
    preds = np.concatenate(preds)

    csv = {'imgs': np.array(image_dir), 'pred': np.array(preds),
           }
    csv = pd.DataFrame(csv)
    print(csv)

    csv.to_csv('results.csv', index=False)

追溯

  return _VF.meshgrid(tensors, **kwargs)  # type: ignore[attr-defined]
Traceback (most recent call last):
  File "/media/cvpr/CM_22/OOD_CV/test.py", line 93, in <module>
    entropy1 = -torch.sum(torch.softmax(pred1[:, :10], dim=1) * torch.logsoftmax(pred1[:, :10], dim=1), dim=-1,
AttributeError: module 'torch' has no attribute 'logsoftmax'

由于 PyTorch 版本冲突，我已更换为最新的 PyTorch 版本，但现在出现模糊错误

  return _VF.meshgrid(tensors, **kwargs)  # type: ignore[attr-defined]
Traceback (most recent call last):
  File "/media/cvpr/CM_22/OOD_CV/test.py", line 93, in <module>
    entropy1 = -torch.sum(torch.softmax(pred1[:, :10], dim=1) * nn.LogSoftmax(pred1[:, :10], dim=1), dim=-1,
TypeError: __init__() got multiple values for argument 'dim'

实施后

nn.LogSoftMax(dim=1)(pred1[:, :10])

追溯

    entropy1 = -torch.sum(torch.softmax(pred1[:, :10], dim=1) * nn.LogSoftmax(dim=1)(pred1[:, :10]), dim=-1, keep_dim=True)
TypeError: sum() received an invalid combination of arguments - got (Tensor, keep_dim=bool, dim=int), but expected one of:
 * (Tensor input, *, torch.dtype dtype)
      didn't match because some of the keywords were incorrect: keep_dim, dim
 * (Tensor input, tuple of ints dim, bool keepdim, *, torch.dtype dtype, Tensor out)
 * (Tensor input, tuple of names dim, bool keepdim, *, torch.dtype dtype, Tensor out)

然后删除keep_dim=True参数

追溯

Traceback (most recent call last):
  File "/media/cvpr/CM_22/OOD_CV/test.py", line 97, in <module>
    pred = torch.softmax(pred1[:, :10], dim=1) * (entropy - entropy1) / entropy + torch.softmax(pred2[:, :10],
RuntimeError: The size of tensor a (10) must match the size of tensor b (128) at non-singleton dimension 1

Answer 1

nn.LogSoftMax 是一个必须先实例化然后调用的模块（这是在执行其forward 方法时）。试试这个：

nn.LogSoftMax(dim=1)(pred1[:, :10])

相反，您还可以使用此模块的功能形式：

torch.nn.functional.log_softmax(pred1[:, :10], dim=1)