[BUG] Issue with ImageNet training set.

## 🐛🐛 Bug Report

### ⚗️ Current Behavior
I tried using ImageNet-1k directly from Active Loop. After validating on PyTorch's pre-trained ResNet-18, I get 82% validation accuracy, which is way too high.

**Input Code**
- REPL or Repo link if applicable:

```python
import deeplake
from PIL import Image
import numpy as np
import os, time
import torch
from torchvision import transforms, models

# Connect to the training and testing datasets
ds_train = deeplake.load("hub://activeloop/imagenet-train", token="my token")
ds_test = deeplake.load("hub://activeloop/imagenet-val", token="my token")

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

def convert_to_rgb(image):
if image.mode != 'RGB':
image = image.convert('RGB')
return image

mean = (0.485, 0.456, 0.406)
std = (0.229, 0.224, 0.225)
tform= transforms.Compose(
[
transforms.Resize(256),
transforms.CenterCrop(224),
transforms.Lambda(convert_to_rgb),
transforms.ToTensor(),
transforms.Normalize(mean, std)
]
)

batch_size = 128

# Since torchvision transforms expect PIL images, we use the 'pil' decode_method for the 'images' tensor. This is much faster than running ToPILImage inside the transform
train_loader = ds_train.pytorch(num_workers = 0, shuffle = True, transform = {'images': tform, 'labels': None}, batch_size = batch_size, decode_method = {'images': 'pil'})
test_loader = ds_test.pytorch(num_workers = 0, transform = {'images': tform, 'labels': None}, batch_size = batch_size, decode_method = {'images': 'pil'})

model = torchvision.models.resnet18(weights="DEFAULT")
device = torch.device("cuda")
model.to(device)
model.eval().cuda() # Needs CUDA, don't bother on CPUs
correct = 0
total = 0
with torch.no_grad():
for x, y in tqdm(test_loader):
y_pred = model(x.cuda())
correct += (y_pred.argmax(axis=1) == y.cuda()).sum().item()
total += len(y)
print(correct / total)
```

**Expected behavior/code**
The ResNet-18 pre-trained model is taken directly from the PyTorch hub. The expected validation accuracy is 69.76% (and I verified this using the Kaggle version of ImageNet). Check this PyTorch link for evidence: https://pytorch.org/vision/main/models/generated/torchvision.models.resnet18.html.
Note: In my transforms, I include a "convert_to_rgb" transform because some of the images from the training and testing sets from the Active Loop hub are grayscale.