import os
import json
import random
from PIL import Image

import torch
from torch.utils.data import Dataset
from torchvision import transforms
from transformers import T5TokenizerFast
from data.transforms import build_coco_transform




class CocoCaptionDataset(Dataset):
    def __init__(
        self,
        split="train",
        image_size=224,
        tokenizer_name="t5-small",
        max_caption_length=64,
        data_dir="data/processed",
        random_caption=True,
        normalize=True,
    ):

        assert split in ["train", "val", "test"]

        self.split = split
        self.image_size = image_size
        self.random_caption = random_caption
        self.max_caption_length = max_caption_length
        self.images_dir = os.path.join(data_dir, "images")
        self.tokenizer = T5TokenizerFast.from_pretrained(tokenizer_name)


        # Load captions.json and splits.json
        captions_file = os.path.join(data_dir, "captions.json")
        splits_file = os.path.join(data_dir, "splits.json")

        with open(captions_file) as f:
            self.captions_data = json.load(f)

        with open(splits_file) as f:
            self.splits = json.load(f)


        # Cast IDs to strings
        self.image_ids = [str(i) for i in self.splits[split]]

        self.transform = build_coco_transform(image_size=image_size)

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

    def __getitem__(self, idx):
        image_id = self.image_ids[idx]

        img_path = os.path.join(self.images_dir, f"{int(image_id):012d}.jpg")
        img = Image.open(img_path).convert("RGB")
        pixel_values = self.transform(img)

        captions = self.captions_data[image_id]["captions"]

        if self.random_caption:
            caption = random.choice(captions)
        else:
            caption = captions[0]  # deterministic for eval

        #  Tokenize caption (no prefix needed for T5 small) 
        encoding = self.tokenizer(
            caption,
            padding="max_length",
            truncation=True,
            max_length=self.max_caption_length,
            return_tensors="pt"
        )

        input_ids = encoding.input_ids.squeeze(0)
        attention_mask = encoding.attention_mask.squeeze(0)

        return {
            "pixel_values": pixel_values,
            "input_ids": input_ids,
            "attention_mask": attention_mask,
            "image_id": image_id,
        }