app.py

import gradio as gr
import base64
import json
import os
import shutil
import uuid
import glob
from huggingface_hub import CommitScheduler, HfApi, snapshot_download
from pathlib import Path
import git
from datasets import Dataset, Features, Value, Sequence, Image as ImageFeature
import threading
import time
from utils import process_and_push_dataset
from datasets import load_dataset

api = HfApi(token=os.environ["HF_TOKEN"])

VALID_DATASET = load_dataset("taesiri/IERv2-Subset-Validation-150", split="train")

VALID_DATASET_POST_IDS = (
    load_dataset(
        "taesiri/IERv2-Subset-Validation-150", split="train", columns=["post_id"]
    )
    .to_pandas()["post_id"]
    .tolist()
)

POST_ID_TO_ID_MAP = {post_id: idx for idx, post_id in enumerate(VALID_DATASET_POST_IDS)}

DATASET_REPO = "taesiri/AIImageEditingResults_Intemediate2"
FINAL_DATASET_REPO = "taesiri/AIImageEditingResults"


# Download existing data from hub
def sync_with_hub():
    """
    Synchronize local data with the hub by cloning the dataset repo
    """
    print("Starting sync with hub...")
    data_dir = Path("./data")
    if data_dir.exists():
        # Backup existing data
        backup_dir = Path("./data_backup")
        if backup_dir.exists():
            shutil.rmtree(backup_dir)
        shutil.copytree(data_dir, backup_dir)

    # Clone/pull latest data from hub
    # Use token in the URL for authentication following HF's new format
    token = os.environ["HF_TOKEN"]
    username = "taesiri"  # Extract from DATASET_REPO
    repo_url = f"https://{username}:{token}@huggingface.co/datasets/{DATASET_REPO}"
    hub_data_dir = Path("hub_data")

    if hub_data_dir.exists():
        # If repo exists, do a git pull
        print("Pulling latest changes...")
        repo = git.Repo(hub_data_dir)
        origin = repo.remotes.origin
        # Set the new URL with token
        if "https://" in origin.url:
            origin.set_url(repo_url)
        origin.pull()
    else:
        # Clone the repo with token
        print("Cloning repository...")
        git.Repo.clone_from(repo_url, hub_data_dir)

    # Merge hub data with local data
    hub_data_source = hub_data_dir / "data"
    if hub_data_source.exists():
        # Create data dir if it doesn't exist
        data_dir.mkdir(exist_ok=True)

        # Copy files from hub
        for item in hub_data_source.glob("*"):
            if item.is_dir():
                dest = data_dir / item.name
                if not dest.exists():  # Only copy if doesn't exist locally
                    shutil.copytree(item, dest)

    # Clean up cloned repo
    if hub_data_dir.exists():
        shutil.rmtree(hub_data_dir)
    print("Finished syncing with hub!")


scheduler = CommitScheduler(
    repo_id=DATASET_REPO,
    repo_type="dataset",
    folder_path="./data",
    path_in_repo="data",
    every=1,
)


def load_question_data(question_id):
    """
    Load a specific question's data
    Returns a tuple of all form fields
    """
    if not question_id:
        return [None] * 11  # Reduced number of fields

    # Extract the ID part before the colon from the dropdown selection
    question_id = (
        question_id.split(":")[0].strip() if ":" in question_id else question_id
    )

    json_path = os.path.join("./data", question_id, "question.json")
    if not os.path.exists(json_path):
        print(f"Question file not found: {json_path}")
        return [None] * 11

    try:
        with open(json_path, "r", encoding="utf-8") as f:
            data = json.loads(f.read().strip())

        # Load images
        def load_image(image_path):
            if not image_path:
                return None
            full_path = os.path.join(
                "./data", question_id, os.path.basename(image_path)
            )
            return full_path if os.path.exists(full_path) else None

        question_images = data.get("question_images", [])
        rationale_images = data.get("rationale_images", [])

        return [
            (
                ",".join(data["question_categories"])
                if isinstance(data["question_categories"], list)
                else data["question_categories"]
            ),
            data["question"],
            data["final_answer"],
            data.get("rationale_text", ""),
            load_image(question_images[0] if question_images else None),
            load_image(question_images[1] if len(question_images) > 1 else None),
            load_image(question_images[2] if len(question_images) > 2 else None),
            load_image(question_images[3] if len(question_images) > 3 else None),
            load_image(rationale_images[0] if rationale_images else None),
            load_image(rationale_images[1] if len(rationale_images) > 1 else None),
            question_id,
        ]
    except Exception as e:
        print(f"Error loading question {question_id}: {str(e)}")
        return [None] * 11


def load_post_image(post_id):
    if not post_id:
        return [
            None
        ] * 33  # source image + instruction + simplified_instruction + 10 triplets

    idx = POST_ID_TO_ID_MAP[post_id]
    source_image = VALID_DATASET[idx]["image"]
    instruction = VALID_DATASET[idx]["instruction"]
    simplified_instruction = VALID_DATASET[idx]["simplified_instruction"]

    # Load existing responses if any
    post_folder = os.path.join("./data", str(post_id))
    metadata_path = os.path.join(post_folder, "metadata.json")

    if os.path.exists(metadata_path):
        with open(metadata_path, "r") as f:
            metadata = json.load(f)

        # Initialize response data
        responses = [(None, "", "")] * 10  # Initialize with empty notes

        # Fill in existing responses
        for response in metadata["responses"]:
            idx = response["response_id"]
            if idx < 10:  # Ensure we don't exceed our UI limit
                image_path = os.path.join(post_folder, response["image_path"])
                responses[idx] = (
                    image_path,
                    response["answer_text"],
                    response.get("notes", ""),
                )

        # Flatten responses for output
        flat_responses = [item for triplet in responses for item in triplet]
        return [source_image, instruction, simplified_instruction] + flat_responses

    # If no existing responses, return source image, instructions and empty responses
    return [source_image, instruction, simplified_instruction] + [None] * 30


def generate_json_files(source_image, responses, post_id):
    """
    Save the source image and multiple responses to the data directory

    Args:
        source_image: Path to the source image
        responses: List of (image, answer, notes) tuples
        post_id: The post ID from the dataset
    """
    # Create parent data folder if it doesn't exist
    parent_data_folder = "./data"
    os.makedirs(parent_data_folder, exist_ok=True)

    # Create/clear post_id folder
    post_folder = os.path.join(parent_data_folder, str(post_id))
    if os.path.exists(post_folder):
        shutil.rmtree(post_folder)
    os.makedirs(post_folder)

    # Save source image
    source_image_path = os.path.join(post_folder, "source_image.png")
    if isinstance(source_image, str):
        shutil.copy2(source_image, source_image_path)
    else:
        gr.processing_utils.save_image(source_image, source_image_path)

    # Create responses data
    responses_data = []
    for idx, (response_image, answer_text, notes) in enumerate(responses):
        if response_image and answer_text:  # Only process if both image and text exist
            response_folder = os.path.join(post_folder, f"response_{idx}")
            os.makedirs(response_folder)

            # Save response image
            response_image_path = os.path.join(response_folder, "response_image.png")
            if isinstance(response_image, str):
                shutil.copy2(response_image, response_image_path)
            else:
                gr.processing_utils.save_image(response_image, response_image_path)

            # Add to responses data
            responses_data.append(
                {
                    "response_id": idx,
                    "answer_text": answer_text,
                    "notes": notes,
                    "image_path": f"response_{idx}/response_image.png",
                }
            )

    # Create metadata JSON
    metadata = {
        "post_id": post_id,
        "source_image": "source_image.png",
        "responses": responses_data,
    }

    # Save metadata
    with open(os.path.join(post_folder, "metadata.json"), "w", encoding="utf-8") as f:
        json.dump(metadata, f, ensure_ascii=False, indent=2)

    return post_folder


def get_statistics():
    """
    Scan the data folder and return statistics about the responses
    """
    data_dir = Path("./data")
    if not data_dir.exists():
        return "No data directory found"

    total_expected_posts = len(VALID_DATASET_POST_IDS)
    processed_post_ids = set()
    posts_with_responses = 0
    total_responses = 0
    responses_per_post = []  # List to track number of responses for each post

    for metadata_file in data_dir.glob("*/metadata.json"):
        post_id = metadata_file.parent.name
        if post_id in VALID_DATASET_POST_IDS:  # Only count valid posts
            processed_post_ids.add(post_id)
            try:
                with open(metadata_file, "r") as f:
                    metadata = json.load(f)
                    num_responses = len(metadata.get("responses", []))
                    responses_per_post.append(num_responses)
                    if num_responses > 0:
                        posts_with_responses += 1
                        total_responses += num_responses
            except:
                continue

    missing_posts = set(map(str, VALID_DATASET_POST_IDS)) - processed_post_ids
    total_processed = len(processed_post_ids)

    # Calculate additional statistics
    if responses_per_post:
        responses_per_post.sort()
        median_responses = responses_per_post[len(responses_per_post) // 2]
        max_responses = max(responses_per_post)
        avg_responses = (
            total_responses / posts_with_responses if posts_with_responses > 0 else 0
        )
    else:
        median_responses = max_responses = avg_responses = 0

    stats = f"""
    📊 Collection Statistics:
    
    Dataset Coverage:
    - Total Expected Posts: {total_expected_posts}
    - Posts Processed: {total_processed}
    - Missing Posts: {len(missing_posts)} ({', '.join(list(missing_posts)[:5])}{'...' if len(missing_posts) > 5 else ''})
    - Coverage Rate: {(total_processed/total_expected_posts*100):.2f}%
    
    Response Statistics:
    - Posts with Responses: {posts_with_responses}
    - Posts without Responses: {total_processed - posts_with_responses}
    - Total Individual Responses: {total_responses}
    
    Response Distribution:
    - Median Responses per Post: {median_responses}
    - Average Responses per Post: {avg_responses:.2f}
    - Maximum Responses for a Post: {max_responses}
    """
    return stats


# Build the Gradio app
with gr.Blocks() as demo:
    gr.Markdown("# Image Response Collector")

    # Source image selection at the top
    with gr.Row():
        with gr.Column():
            post_id_dropdown = gr.Dropdown(
                label="Select Post ID to Load Image",
                choices=VALID_DATASET_POST_IDS,
                type="value",
                allow_custom_value=False,
            )
            instruction_text = gr.Textbox(label="Instruction", interactive=False)
            simplified_instruction_text = gr.Textbox(
                label="Simplified Instruction", interactive=False
            )
        source_image = gr.Image(label="Source Image", type="filepath", height=300)

    # Responses in tabs
    with gr.Tabs() as response_tabs:
        responses = []
        for i in range(10):
            with gr.Tab(f"Response {i+1}"):
                img = gr.Image(
                    label=f"Response Image {i+1}", type="filepath", height=300
                )
                txt = gr.Textbox(label=f"Model Name {i+1}", lines=2)
                notes = gr.Textbox(label=f"Miscellaneous Notes {i+1}", lines=3)
                responses.append((img, txt, notes))

    with gr.Row():
        submit_btn = gr.Button("Submit All Responses")
        clear_btn = gr.Button("Clear Form")

    # Add statistics accordion
    with gr.Accordion("Collection Statistics", open=False):
        stats_text = gr.Markdown("Loading statistics...")
        refresh_stats_btn = gr.Button("Refresh Statistics")

        def update_stats():
            return get_statistics()

        refresh_stats_btn.click(fn=update_stats, outputs=[stats_text])

        # Move the load event inside the Blocks context
        demo.load(
            fn=get_statistics,
            outputs=[stats_text],
        )

    def submit_responses(
        source_img, post_id, instruction, simplified_instruction, *response_data
    ):
        if not source_img:
            gr.Warning("Please select a source image first!")
            return

        if not post_id:
            gr.Warning("Please select a post ID first!")
            return

        # Convert flat response_data into triplets of (image, text, notes)
        response_triplets = list(
            zip(response_data[::3], response_data[1::3], response_data[2::3])
        )

        # Check for responses with images but no model names
        incomplete_responses = [
            i + 1
            for i, (img, txt, _) in enumerate(response_triplets)
            if img is not None and not txt.strip()
        ]

        if incomplete_responses:
            gr.Warning(
                f"Please provide model names for responses: {', '.join(map(str, incomplete_responses))}!"
            )
            return

        # Filter out empty responses (where both image and model name are empty)
        valid_responses = [
            (img, txt, notes)
            for img, txt, notes in response_triplets
            if img is not None and txt.strip()
        ]

        if not valid_responses:
            gr.Warning("Please provide at least one response (image + model name)!")
            return

        # Generate JSON files with the valid responses
        generate_json_files(source_img, valid_responses, post_id)
        gr.Info("Responses saved successfully! 🎉")

    def clear_form():
        outputs = [None] * (
            1 + 2 + 30
        )  # source image + 2 instruction fields + 10 triplets
        return outputs

    # Connect components
    post_id_dropdown.change(
        fn=load_post_image,
        inputs=[post_id_dropdown],
        outputs=[source_image, instruction_text, simplified_instruction_text]
        + [comp for triplet in responses for comp in triplet],
    )

    submit_inputs = [
        source_image,
        post_id_dropdown,
        instruction_text,
        simplified_instruction_text,
    ] + [comp for triplet in responses for comp in triplet]
    submit_btn.click(fn=submit_responses, inputs=submit_inputs)

    clear_outputs = [source_image, instruction_text, simplified_instruction_text] + [
        comp for triplet in responses for comp in triplet
    ]
    clear_btn.click(fn=clear_form, outputs=clear_outputs)


def process_thread():
    while True:
        try:
            pass
            # process_and_push_dataset(
            #     "./data",
            #     FINAL_DATASET_REPO,
            #     token=os.environ["HF_TOKEN"],
            #     private=True,
            # )
        except Exception as e:
            print(f"Error in process thread: {e}")
        time.sleep(120)  # Sleep for 2 minutes


if __name__ == "__main__":
    print("Initializing app...")
    sync_with_hub()  # Sync before launching the app
    print("Starting Gradio interface...")

    # Start the processing thread when the app starts
    processing_thread = threading.Thread(target=process_thread, daemon=True)
    processing_thread.start()

    demo.launch()