make_flickr_ds.py

import json
import os
from tqdm import tqdm

from datasets import Dataset, load_dataset, Image
import pandas as pd


def load_jsonl(file_path):
    """
    Loads a JSONL file and returns a list of Python dictionaries.
    Each dictionary represents a JSON object from a line in the file.
    """
    data = []
    with open(file_path, 'r', encoding='utf-8') as f:
        for line in f:
            try:
                # Parse each line as a JSON object
                json_object = json.loads(line.strip())
                data.append(json_object)
            except json.JSONDecodeError as e:
                print(f"Error decoding JSON on line: {line.strip()} - {e}")
    return data


def main():
    dsets = ["train", "val", "test"]  # test val
    workdir = "./flickr30k"

    # Load annotations
    annot_fn = os.path.join(workdir, "results.csv")
    df = pd.read_csv(annot_fn, delimiter="|")
    df = pd.DataFrame(df)

    datadict = {}
    for _, row in df.iterrows():
        idx = row["image_name"].replace(".jpg", "")
        if idx not in datadict:
            datadict[idx] = {
                "image_name": row["image_name"],
                "image": os.path.join(workdir, "flickr30k_images", row["image_name"]),
                "sentids": [],
                "split": None,
                "caption": [],
                "narratives": []
            }

        datadict[idx]["sentids"].append(row[" comment_number"])
        datadict[idx]["caption"].append(row[" comment"])

    # Align to narratives splits.
    for split in dsets:
        narr = load_jsonl(os.path.join(workdir, "narratives", f"flickr30k_{split}_captions.jsonl"))
        for item in narr:
            idx = item["image_id"]
            datadict[idx]["split"] = split
            datadict[idx]["narratives"].append(item["caption"])

    # make datasets
    for split in dsets:
        df = pd.DataFrame.from_dict(datadict, orient="index")
        df = df[df["split"] == split]
        ds = Dataset.from_pandas(df)
        ds = ds.remove_columns(["__index_level_0__", "split"])
        ds = ds.cast_column("image", Image())
        ds.save_to_disk(os.path.join(workdir, "datasets", "data", split), max_shard_size="400MB")

    return


def test_dataset():
    ds = load_dataset("./flickr30k/datasets")  # , split="val"
    print(ds["train"][0])
    

if __name__ == "__main__":
    # main()
    test_dataset()