6 channel input with Imagefolder up

1

I have two separate datasets. Each set has 3224224 images. I want to input those datasets into a network as a 6 channel input. Both datasets have the same labels. I want to use Imagefolder for this.

trainset_cort = torchvision.datasets.ImageFolder(root=data_dir_train_cort,
                                transform=transform)

                        
                                          
trainset_slice = torchvision.datasets.ImageFolder(root=data_dir_train_slice,
                                transform=transform)
                                
  

      

trainloader= torch.utils.data.DataLoader(trainset_new
             , 
             batch_size=4,shuffle=True, num_workers=2)



testset_cort = torchvision.datasets.ImageFolder(root=data_dir_val_cort,
                                transform=transform)

                                          
testset_slice = torchvision.datasets.ImageFolder(root=data_dir_val_slice,
                                                 transform=transform)
                                                 


testloader= torch.utils.data.DataLoader(testset_new
             , batch_size=4,shuffle=True, num_workers=2)

I’m trying above code. i want to create both trainset_new and testset_new as 6 by 224 by 224 imageset

2

You could create your own Dataset and concatenate both images:

class MyDataset(Dataset):
    def __init__(self, path_cort, path_slice, transform=None):
        self.data_cort = datasets.ImageFolder(
            root=path_cort, transform=transform)
        self.data_slice = datasets.ImageFolder(
            root=path_slice, transform=transform)
        
    def __getitem__(self, index):
        x_cort, y = self.data_cort[index]
        x_slice, _ = self.data_slice[index]
        x = torch.cat((x_cort, x_slice), dim=1)
        return x, y
    
    def __len__(self):
        return len(self.data_cort) # assert both datasets have equal length
3

Thank you. I’ll try this method.

4

using this I get the error as:

module.__init__() takes at most 2 arguments (3 given)

I am using google colab for this and the code I ran is this:

import torch
import torch.nn as nn
import torch.optim as optim
from torch.optim import lr_scheduler
import numpy as np
import torchvision
from torchvision import datasets, models, transforms
import matplotlib.pyplot as plt
import time
import os
import copy

class CustomDataset(datasets):
    def __init__(self, path_cort, path_slice, transform):
        self.data_cort = datasets.ImageFolder(
            root=path_cort, transform=transform)
        self.data_slice = datasets.ImageFolder(
            root=path_slice, transform=transform)
        
    def __getitem__(self, index):
        x_cort, y = self.data_cort[index]
        x_slice, _ = self.data_slice[index]
        x = torch.cat((x_cort, x_slice), dim=1)
        return x, y
    
    def __len__(self):
        return len(self.data_cort) # assert both datasets have equal length

root = "/content/drive/MyDrive/Rohit/Human Pose Estimation/UTD MHAD/"
transform = transforms.Compose([
                                transforms.Resize((224,224)),
                                transforms.ToTensor(),
                                transforms.Normalize(mean, std)
                                ])
Dataset = CustomDataset(root+"angle/train", root+"anglevel/train", transform)

num_classes = 27

Inside the respective train folders, each class has several images.

Hope you will help me out, thanks in advance

5

Could you device the CustomDataset from torch.utils.data.Dataset instead of torchvision.datasets and check, if it would solve the issue?