how do I use ragavsachdeva/popmanga_test to train Dinov2 i am doing this on kaggle here code
from transformers import Dinov2Config, Dinov2ForImageClassification, Trainer, TrainingArguments
from datasets import load_dataset
from PIL import Image
import numpy as np
import torch
from torch.utils.data import Dataset

Preprocess images

def read_image(path_to_image):
with open(path_to_image, "rb") as file:
image = Image.open(file).convert("RGB")
image = np.array(image)
return image

Custom dataset class

class MangaDataset(Dataset):
def init(self, dataset_split, target_size=(224, 224)): # Specify target size
self.dataset_split = dataset_split
self.target_size = target_size # Target size for resizing

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

def __getitem__(self, idx):
    example = self.dataset_split[idx]
    image_path = example["image_path"]
    text_classification = example.get("text_classification", [])

    # Check if text_classification is not empty
    if not text_classification:
        # Handle missing labels gracefully
        print(f"No classification labels found for index {idx}. Assigning default label.")
        label = -1  # Or some other default value indicating no label
    else:
        label = text_classification[0]  # Select the first label if it's multi-class

    image = read_image(image_path)

    # Resize image to target size
    image = Image.fromarray(image)  # Convert numpy array back to PIL Image for resizing
    image = image.resize(self.target_size, Image.LANCZOS)  # Resize with high-quality resampling

    image_tensor = torch.tensor(np.array(image), dtype=torch.float32).permute(2, 0, 1) / 255.0
    return {"pixel_values": image_tensor, "label": torch.tensor(label, dtype=torch.long)}

Main script

if name == "main":
# Load the dataset
dataset = load_dataset("ragavsachdeva/popmanga_test", trust_remote_code=True)

# Preprocess splits
train_dataset = MangaDataset(dataset["seen"], target_size=(224, 224))
eval_dataset = MangaDataset(dataset["unseen"], target_size=(224, 224))

# Debugging: Check for missing labels in the training set
for idx in range(len(train_dataset)):
    example = train_dataset[idx]
    if example['label'].item() == -1:  # Assuming -1 is your default value for missing labels
        print(f"Missing label at index {idx}: {example}")

config = Dinov2Config.from_pretrained("facebook/dinov2-base")
model = Dinov2ForImageClassification.from_pretrained("facebook/dinov2-base", config=config)

# Define training arguments
training_args = TrainingArguments(
    output_dir='./results',
    per_device_train_batch_size=8,
    per_device_eval_batch_size=8,
    num_train_epochs=3,
    learning_rate=5e-5,
    eval_strategy="epoch",
    save_strategy="epoch",
    logging_dir='./logs',
    logging_steps=10,
)

# Initialize Trainer
trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=train_dataset,
    eval_dataset=eval_dataset,
)

# Start training
trainer.train()