first commit

app.py

@@ -0,0 +1,729 @@
+import os
+import sys
+import glob
+import json
+import random
+import re
+from functools import partial
+from datetime import datetime
+from collections import defaultdict, Counter
+
+import gradio as gr
+from loguru import logger
+
+# --- Global State (unchanged) ---
+# --- Global State (unchanged) ---
+GLOBAL_STATE = {
+    "participant_id": None,
+    "data_loaded": False,
+    "all_eval_data": [],
+    "shuffled_indices": [],
+    "current_prompt_index": 0,
+    "current_criterion_index": 0,
+    "image_mapping": {},
+    "image_dir": "",
+    "evaluation_results": {},
+    "image_orders": {},
+    "start_time": None,
+    "end_time": None,
+    "current_ranks": {},
+    "current_absolute_score": None,
+    # ▼▼▼ 追加 ▼▼▼
+    "current_absolute_score_worst": None,
+}
+
+# --- Configuration (unchanged) ---
+BASE_RESULTS_DIR = "./results"
+LOG_DIR = "./logs"
+COMBINED_DATA_DIR = "./combined_data"
+IMAGE_SUBDIR = os.path.join("lapwing", "images")
+MAPPING_FILENAME = "combination_to_filename.json"
+CONDITIONS = ["Ours", "w_o_Proto_Loss", "w_o_HitL", "w_o_Tuning", "LLM-based"]
+CRITERIA = ["Alignment", "Naturalness", "Attractiveness"]
+
+CRITERIA_GUIDANCE_JP = [
+    "テキストと表情がどれだけ一致しているか",
+    "テキストの感情に沿ったセリフを言っていると想像したとき、表情がどれだけ自然か",
+    "テキストの感情に沿ったセリフを言っていると想像したとき、表情がどれだけ魅力的か"
+]
+CRITERIA_GUIDANCE_EN = [
+    "how well the expression aligns with the text",
+    "imagining the character is speaking a line that matches the emotion of the text, how natural the facial expression is",
+    "imagining the character is speaking a line that matches the emotion of the text, how attractive the facial expression is"
+]
+IMAGE_LABELS = ['A', 'B', 'C', 'D', 'E']
+
+
+# --- Helper Functions ---
+def get_image_path_from_prediction(prediction: dict) -> str:
+    if not GLOBAL_STATE["image_mapping"]:
+        logger.error("Image mapping is not loaded.")
+        return ""
+    indices = prediction.get("blendshape_index", {})
+    if not isinstance(indices, dict):
+        logger.error(f"blendshape_index is not a dictionary: {indices}")
+        return ""
+    sorted_indices = sorted(indices.items(), key=lambda item: int(item[0]))
+    key = ",".join(str(idx) for _, idx in sorted_indices)
+    filename = GLOBAL_STATE["image_mapping"].get(key)
+    if not filename:
+        logger.warning(f"No image found for blendshape key: {key}")
+        return ""
+    return os.path.join(GLOBAL_STATE["image_dir"], filename)
+
+
+# ▼▼▼ 2. prompt_categoryを読み込むように修正 ▼▼▼
+def load_evaluation_data(participant_id: str):
+    mapping_path = os.path.join(COMBINED_DATA_DIR, MAPPING_FILENAME)
+    if not os.path.exists(mapping_path):
+        return f"<p class='feedback red'>Error: Mapping file not found at {mapping_path}</p>", gr.update(
+            interactive=True), gr.update(interactive=False)
+
+    with open(mapping_path, 'r', encoding='utf-8') as f:
+        GLOBAL_STATE["image_mapping"] = json.load(f)["mapping"]
+    GLOBAL_STATE["image_dir"] = os.path.join(COMBINED_DATA_DIR, IMAGE_SUBDIR)
+    logger.info(f"Successfully loaded image mapping. Image directory: {GLOBAL_STATE['image_dir']}")
+
+    participant_dir = os.path.join(BASE_RESULTS_DIR, participant_id)
+    if not os.path.isdir(participant_dir):
+        return f"<p class='feedback red'>Error: Participant directory not found: {participant_dir}</p>", gr.update(
+            interactive=True), gr.update(interactive=False)
+
+    merged_data = defaultdict(lambda: {"predictions": {}, "category": None})
+    found_files = 0
+    for cond in CONDITIONS:
+        cond_dir = os.path.join(participant_dir, cond)
+        pattern = os.path.join(cond_dir, f"{participant_id}_{cond}_*.jsonl")
+        files = glob.glob(pattern)
+        if not files:
+            logger.warning(f"No prediction file found for condition '{cond}' with pattern: {pattern}")
+            continue
+        found_files += 1
+        with open(files[0], 'r', encoding='utf-8') as f:
+            for line in f:
+                data = json.loads(line)
+                prompt = data["text_prompt"]
+                merged_data[prompt]["predictions"][cond] = data["prediction"]
+                if not merged_data[prompt]["category"]:
+                    merged_data[prompt]["category"] = data.get("prompt_category")
+
+    if found_files != len(CONDITIONS):
+        return f"<p class='feedback red'>Error: Found prediction files for only {found_files}/{len(CONDITIONS)} conditions.</p>", gr.update(
+            interactive=True), gr.update(interactive=False)
+
+    GLOBAL_STATE["all_eval_data"] = [
+        {"prompt": p, "predictions": d["predictions"], "category": d["category"]}
+        for p, d in merged_data.items() if len(d["predictions"]) == len(CONDITIONS)
+    ]
+    # ▲▲▲ END OF UPDATE ▲▲▲
+
+    if not GLOBAL_STATE["all_eval_data"]:
+        return "<p class='feedback red'>Error: No valid evaluation data could be loaded.</p>", gr.update(
+            interactive=True), gr.update(interactive=False)
+
+    GLOBAL_STATE["shuffled_indices"] = list(range(len(GLOBAL_STATE["all_eval_data"])))
+    random.shuffle(GLOBAL_STATE["shuffled_indices"])
+    GLOBAL_STATE["current_prompt_index"] = 0
+    GLOBAL_STATE["current_criterion_index"] = 0
+    GLOBAL_STATE["data_loaded"] = True
+    GLOBAL_STATE["start_time"] = datetime.now()
+    for i in range(len(GLOBAL_STATE["all_eval_data"])):
+        prompt_text = GLOBAL_STATE["all_eval_data"][i]["prompt"]
+        GLOBAL_STATE["evaluation_results"][prompt_text] = {}
+
+    logger.info(f"Loaded and merged data for {len(GLOBAL_STATE['all_eval_data'])} prompts.")
+    done_msg = "<p class='feedback green'>Data loaded successfully. Please proceed to the 'Evaluation' tab. / データの読み込みに成功しました。「評価」タブに進んでください。</p>"
+    return done_msg, gr.update(interactive=False, visible=False), gr.update(interactive=True)
+
+
+# --- Core Logic ---
+
+def _create_button_updates():
+    updates = []
+    for img_label in IMAGE_LABELS:
+        selected_rank = GLOBAL_STATE["current_ranks"].get(img_label)
+        for rank_val in range(1, 6):
+            if rank_val == selected_rank:
+                updates.append(gr.update(variant='primary'))
+            else:
+                updates.append(gr.update(variant='secondary'))
+    return updates
+
+
+def handle_rank_button_click(image_label, rank):
+    if GLOBAL_STATE["current_ranks"].get(image_label) == rank:
+        GLOBAL_STATE["current_ranks"][image_label] = None
+    else:
+        GLOBAL_STATE["current_ranks"][image_label] = rank
+    return _create_button_updates()
+
+
+def handle_absolute_score_click(score):
+    if GLOBAL_STATE["current_absolute_score"] == score:
+        GLOBAL_STATE["current_absolute_score"] = None
+    else:
+        GLOBAL_STATE["current_absolute_score"] = score
+
+    updates = []
+    for i in range(1, 8):
+        if i == GLOBAL_STATE["current_absolute_score"]:
+            updates.append(gr.update(variant='primary'))
+        else:
+            updates.append(gr.update(variant='secondary'))
+    return updates
+
+
+# ▼▼▼ 追加 ▼▼▼
+def handle_absolute_score_worst_click(score):
+    if GLOBAL_STATE["current_absolute_score_worst"] == score:
+        GLOBAL_STATE["current_absolute_score_worst"] = None
+    else:
+        GLOBAL_STATE["current_absolute_score_worst"] = score
+
+    updates = []
+    for i in range(1, 8):
+        if i == GLOBAL_STATE["current_absolute_score_worst"]:
+            updates.append(gr.update(variant='primary'))
+        else:
+            updates.append(gr.update(variant='secondary'))
+    return updates
+
+# ▼▼▼ 1. UIフリーズ問題を修正 ▼▼▼
+
+
+# ▼▼▼ 修正後の display_current_prompt_and_criterion 関数 ▼▼▼
+def display_current_prompt_and_criterion():
+    if not GLOBAL_STATE["data_loaded"] or GLOBAL_STATE["current_prompt_index"] >= len(GLOBAL_STATE["all_eval_data"]):
+        done_msg = "<p class='feedback green' style='text-align: center; font-size: 1.2em;'>All prompts have been evaluated! Please proceed to the 'Export' tab. <br>すべてのプロンプトの評価が完了しました！「エクスポート」タブに進んでください。</p>"
+        empty_button_updates = [gr.update(variant='secondary')] * 25
+        empty_abs_updates = [gr.update(variant='secondary')] * 7
+        return [
+            gr.update(value="Finished! / 完了!"),
+            gr.update(value=""),
+            gr.update(value=done_msg),
+            gr.update(value="", visible=False),
+            *[gr.update(value=None)] * 5,
+            *empty_button_updates,
+            gr.update(visible=False),  # abs_group_best
+            *empty_abs_updates,
+            gr.update(visible=False),  # abs_group_worst
+            *empty_abs_updates,
+            gr.update(interactive=False),
+            gr.update(interactive=False)
+        ]
+
+    prompt_idx = GLOBAL_STATE["shuffled_indices"][GLOBAL_STATE["current_prompt_index"]]
+    criterion_idx = GLOBAL_STATE["current_criterion_index"]
+
+    current_data = GLOBAL_STATE["all_eval_data"][prompt_idx]
+    prompt_text = current_data["prompt"]
+    criterion_name = CRITERIA[criterion_idx]
+
+    progress_text = f"Prompt {GLOBAL_STATE['current_prompt_index'] + 1} / {len(GLOBAL_STATE['all_eval_data'])}  -  **{criterion_name}**"
+    prompt_display_text = f"## \"{prompt_text}\""
+    guidance_text = f"### Please rank the 5 images based on **{CRITERIA_GUIDANCE_EN[criterion_idx]}**.<br>5つの画像を、**「{CRITERIA_GUIDANCE_JP[criterion_idx]}」**を基準にランキング付けしてください。"
+
+    if criterion_idx == 0:
+        GLOBAL_STATE["image_orders"] = {}
+
+    if criterion_name not in GLOBAL_STATE["image_orders"]:
+        conditions_shuffled = random.sample(CONDITIONS, len(CONDITIONS))
+        GLOBAL_STATE["image_orders"][criterion_name] = conditions_shuffled
+
+    current_image_order = GLOBAL_STATE["image_orders"][criterion_name]
+    image_updates = []
+    for cond_name in current_image_order:
+        prediction = current_data["predictions"][cond_name]
+        img_path = get_image_path_from_prediction(prediction)
+        image_updates.append(gr.update(value=img_path if img_path and os.path.exists(img_path) else None))
+
+    saved_ranks_dict = GLOBAL_STATE["evaluation_results"].get(prompt_text, {}).get("ranks", {}).get(criterion_name)
+    if saved_ranks_dict:
+        label_to_condition = {label: cond for label, cond in zip(IMAGE_LABELS, current_image_order)}
+        condition_to_label = {v: k for k, v in label_to_condition.items()}
+        GLOBAL_STATE["current_ranks"] = {
+            condition_to_label[cond]: rank for cond, rank in saved_ranks_dict.items() if cond in condition_to_label
+        }
+    else:
+        GLOBAL_STATE["current_ranks"] = {label: None for label in IMAGE_LABELS}
+
+    button_updates = _create_button_updates()
+
+    # --- Absolute Score (Best) ---
+    is_alignment_criterion = (criterion_name == "Alignment")
+    abs_group_update = gr.update(visible=is_alignment_criterion)
+    saved_abs_score = GLOBAL_STATE["evaluation_results"].get(prompt_text, {}).get("absolute_score")
+    GLOBAL_STATE["current_absolute_score"] = saved_abs_score if is_alignment_criterion else None
+
+    abs_button_updates = []
+    for i in range(1, 8):
+        variant = 'primary' if i == GLOBAL_STATE["current_absolute_score"] else 'secondary'
+        abs_button_updates.append(gr.update(variant=variant))
+
+    # --- Absolute Score (Worst) ---
+    abs_group_worst_update = gr.update(visible=is_alignment_criterion)
+    saved_abs_score_worst = GLOBAL_STATE["evaluation_results"].get(prompt_text, {}).get("absolute_score_worst")
+    GLOBAL_STATE["current_absolute_score_worst"] = saved_abs_score_worst if is_alignment_criterion else None
+
+    abs_button_worst_updates = []
+    for i in range(1, 8):
+        variant = 'primary' if i == GLOBAL_STATE["current_absolute_score_worst"] else 'secondary'
+        abs_button_worst_updates.append(gr.update(variant=variant))
+
+    return [
+        gr.update(value=progress_text),
+        gr.update(value=prompt_display_text),
+        gr.update(value=guidance_text),
+        gr.update(value="", visible=False),
+        *image_updates,
+        *button_updates,
+        abs_group_update,
+        *abs_button_updates,
+        abs_group_worst_update,
+        *abs_button_worst_updates,
+        gr.update(
+            interactive=(GLOBAL_STATE["current_prompt_index"] > 0 or GLOBAL_STATE["current_criterion_index"] > 0)),
+        gr.update(interactive=True)
+    ]
+
+
+# ▼▼▼ 修正後の validate_and_navigate 関数 ▼▼▼
+def validate_and_navigate():
+    ranks = GLOBAL_STATE["current_ranks"]
+    error_msg = None
+    criterion_name = CRITERIA[GLOBAL_STATE["current_criterion_index"]]
+    is_alignment_criterion = (criterion_name == "Alignment")
+
+    # --- Validation ---
+    if any(r is None for r in ranks.values()):
+        error_msg = "Please rank all 5 images. / 5つすべての画像を評価してください。"
+    elif 1 not in ranks.values():
+        error_msg = "You must assign a rank of '1' to at least one image. / 最低1つは「1位」を付けてください。"
+    elif is_alignment_criterion and GLOBAL_STATE["current_absolute_score"] is None:
+        error_msg = "Please provide an absolute score for the BEST matching image (1-7). / 最も一致している画像について、絶対評価（1～7）を選択してください。"
+    elif is_alignment_criterion and GLOBAL_STATE["current_absolute_score_worst"] is None:
+        error_msg = "Please provide an absolute score for the WORST matching image (1-7). / 最も一致していない画像について、絶対評価（1～7）を選択してください。"
+
+    if error_msg:
+        # The number of components to update is now 53 (1 tab + 52 eval components)
+        no_change_updates = [gr.update()] * 53
+        no_change_updates[4] = gr.update(  # error_display is the 5th component (index 4)
+            value=f"<p class='feedback red' style='font-size: 1.2em; text-align: center;'>{error_msg}</p>",
+            visible=True)
+        return no_change_updates
+
+    # ... (Rank tie-breaking validation logic is unchanged) ...
+    sorted_ranks = sorted(list(ranks.values()))
+    rank_counts = Counter(sorted_ranks)
+    i = 0
+    while i < len(sorted_ranks):
+        current_rank = sorted_ranks[i]
+        count = rank_counts[current_rank]
+        if i + count < len(sorted_ranks):
+            next_rank = sorted_ranks[i + count]
+            expected_next_rank = current_rank + count
+            if next_rank < expected_next_rank:
+                error_msg = f"Ranking rule violation (tie-breaking). After {count} instance(s) of rank '{current_rank}', the next rank must be >= {expected_next_rank}, but it is '{next_rank}'. / 順位付けのルール違反です。'{current_rank}'位が{count}つあるため、次の順位は{expected_next_rank}位以上である必要がありますが、'{next_rank}'位が入力されています。"
+                break
+        i += count
+    if error_msg:
+        no_change_updates = [gr.update()] * 53
+        no_change_updates[4] = gr.update(
+            value=f"<p class='feedback red' style='font-size: 1.2em; text-align: center;'>{error_msg}</p>",
+            visible=True)
+        return no_change_updates
+    # --- End of Validation ---
+
+    prompt_idx = GLOBAL_STATE["shuffled_indices"][GLOBAL_STATE["current_prompt_index"]]
+    current_data = GLOBAL_STATE["all_eval_data"][prompt_idx]
+    prompt_text = current_data["prompt"]
+    current_image_order = GLOBAL_STATE["image_orders"][criterion_name]
+
+    label_to_condition = {label: cond for label, cond in zip(IMAGE_LABELS, current_image_order)}
+    ranks_by_condition = {label_to_condition[label]: rank for label, rank in ranks.items()}
+
+    if "ranks" not in GLOBAL_STATE["evaluation_results"][prompt_text]:
+        GLOBAL_STATE["evaluation_results"][prompt_text]["ranks"] = {}
+    if "orders" not in GLOBAL_STATE["evaluation_results"][prompt_text]:
+        GLOBAL_STATE["evaluation_results"][prompt_text]["orders"] = {}
+
+    GLOBAL_STATE["evaluation_results"][prompt_text]["ranks"][criterion_name] = ranks_by_condition
+    GLOBAL_STATE["evaluation_results"][prompt_text]["orders"][criterion_name] = current_image_order
+
+    if is_alignment_criterion:
+        GLOBAL_STATE["evaluation_results"][prompt_text]["absolute_score"] = GLOBAL_STATE["current_absolute_score"]
+        GLOBAL_STATE["evaluation_results"][prompt_text]["absolute_score_worst"] = GLOBAL_STATE[
+            "current_absolute_score_worst"]
+
+    logger.info(
+        f"Saved rank for P:{GLOBAL_STATE['participant_id']}, Prompt:'{prompt_text}', Criterion:{criterion_name}, Ranks:{ranks_by_condition}")
+
+    GLOBAL_STATE["current_criterion_index"] += 1
+    if GLOBAL_STATE["current_criterion_index"] >= len(CRITERIA):
+        GLOBAL_STATE["current_criterion_index"] = 0
+        GLOBAL_STATE["current_prompt_index"] += 1
+
+    if GLOBAL_STATE["current_prompt_index"] >= len(GLOBAL_STATE["all_eval_data"]):
+        GLOBAL_STATE["end_time"] = datetime.now()
+        eval_panel_updates = display_current_prompt_and_criterion()
+        # Activate export tab on completion
+        return [gr.update(interactive=True)] + eval_panel_updates
+    else:
+        # Keep export tab state as is
+        return [gr.update()] + display_current_prompt_and_criterion()
+
+
+
+def navigate_previous():
+    GLOBAL_STATE["current_criterion_index"] -= 1
+    if GLOBAL_STATE["current_criterion_index"] < 0:
+        GLOBAL_STATE["current_criterion_index"] = len(CRITERIA) - 1
+        GLOBAL_STATE["current_prompt_index"] -= 1
+    GLOBAL_STATE["current_prompt_index"] = max(0, GLOBAL_STATE["current_prompt_index"])
+    return display_current_prompt_and_criterion()
+
+# ▼▼▼ 修正後の export_results 関数 ▼▼▼
+def export_results(participant_id, alignment_reason, naturalness_reason, attractiveness_reason, optional_comment):
+    if not alignment_reason.strip() or not naturalness_reason.strip() or not attractiveness_reason.strip():
+        error_msg = "<p class='feedback red'>Please fill in the reasoning for all three criteria (Alignment, Naturalness, Attractiveness). / 3つの評価基準（一致度, 自然さ, 魅力度）すべての判断理由を記入してください。</p>"
+        return None, error_msg
+
+    if not participant_id:
+        return None, "<p class='feedback red'>Participant ID is missing. / 参加者IDがありません。</p>"
+
+    output_dir = os.path.join(BASE_RESULTS_DIR, participant_id)
+    os.makedirs(output_dir, exist_ok=True)
+    filename = f"evaluation_results_{participant_id}_{datetime.now().strftime('%Y%m%d_%H%M%S')}.json"
+    filepath = os.path.join(output_dir, filename)
+
+    duration = (GLOBAL_STATE["end_time"] - GLOBAL_STATE["start_time"]).total_seconds() if GLOBAL_STATE.get(
+        "start_time") and GLOBAL_STATE.get("end_time") else None
+
+    prompt_to_category = {item["prompt"]: item["category"] for item in GLOBAL_STATE["all_eval_data"]}
+
+    final_results_list = []
+    for prompt, data in GLOBAL_STATE["evaluation_results"].items():
+        if not data: continue
+
+        ranks_data = data.get("ranks", {})
+        orders_data = data.get("orders", {})
+
+        final_results_list.append({
+            "prompt": prompt,
+            "prompt_category": prompt_to_category.get(prompt),
+            "image_order_alignment": orders_data.get("Alignment", []),
+            "image_order_naturalness": orders_data.get("Naturalness", []),
+            "image_order_attractiveness": orders_data.get("Attractiveness", []),
+            "alignment_ranks": ranks_data.get("Alignment", {}),
+            "naturalness_ranks": ranks_data.get("Naturalness", {}),
+            "attractiveness_ranks": ranks_data.get("Attractiveness", {}),
+            "alignment_absolute_score": data.get("absolute_score"),
+            # ▼▼▼ 追加 ▼▼▼
+            "alignment_absolute_score_worst": data.get("absolute_score_worst")
+        })
+
+    export_data = {
+        "metadata": {
+            "participant_id": participant_id,
+            "export_timestamp": datetime.now().isoformat(),
+            "total_prompts_evaluated": len(final_results_list),
+            "evaluation_duration_seconds": duration,
+            "reasoning": {
+                "alignment": alignment_reason,
+                "naturalness": naturalness_reason,
+                "attractiveness": attractiveness_reason,
+            },
+            "optional_comment": optional_comment,
+        },
+        "results": final_results_list
+    }
+
+    try:
+        with open(filepath, 'w', encoding='utf-8') as f:
+            json.dump(export_data, f, ensure_ascii=False, indent=2)
+        logger.info(f"Successfully exported results to: {filepath}")
+    except Exception as e:
+        logger.error(f"Failed to write export file: {e}")
+        return None, f"<p class='feedback red'>An error occurred during file export: {e}</p>"
+
+    upload_link = "https://drive.google.com/drive/folders/1ujIPF-67Y6OG8qBm1TYG3FsmuYxqSAcR?usp=drive_link"
+    status_message = f"""
+    <div class='feedback green' style='text-align: left;'>
+    <p><b>エクスポートが完了しました。/ Export complete.</b></p>
+    <p>上のボタンからJSONファイルをダウンロードし、指定された場所にアップロードして実験を終了してください。ご協力ありがとうございました。</p>
+    <p>Please download the JSON file and upload it to the designated location. Thank you for your cooperation.</p>
+    <p><b>アップロード先 / Upload to:</b> <a href='{upload_link}' target='_blank'>{upload_link}</a></p>
+    </div>"""
+    return gr.update(value=filepath, visible=True), status_message
+
+
+## ▼▼▼ 修正後の create_gradio_interface 関数 ▼▼▼
+def create_gradio_interface():
+    css = """
+    .gradio-container { font-family: 'Arial', sans-serif; }
+    .feedback { padding: 10px; border-radius: 5px; font-weight: bold; text-align: center; margin-top: 10px; }
+    .feedback.green { background-color: #e6ffed; color: #2f6f4a; }
+    .feedback.red { background-color: #ffe6e6; color: #b30000; }
+    .image-label { font-size: 2.5em; font-weight: bold; margin-bottom: 10px; color: #333; }
+    .prompt-display { text-align: center; margin-bottom: 5px; padding: 15px; background-color: #f0f8ff; border-radius: 8px; }
+    .prompt-sub-guidance { text-align: center; font-size: 0.9em; color: #555; margin-top: 5px; margin-bottom: 15px; }
+    .rank-instruction { 
+        color: #D32F2F; 
+        font-size: 1.1em; 
+        text-align: left; 
+        margin-bottom: 20px; 
+        padding: 15px; 
+        border: 1px solid #f5c6cb; 
+        border-radius: 8px;
+        background-color: #f8d7da; 
+        line-height: 1.6;
+    }
+    .rank-instruction ul { padding-left: 20px; margin: 0; }
+    .rank-guidance { text-align: center; margin-bottom: 10px; font-size: 1.2em; }
+    .rank-btn-row { justify-content: center; gap: 5px !important; }
+    .rank-btn { 
+        min-width: 45px !important; 
+        max-width: 45px !important;
+        height: 45px !important;
+        font-size: 1.2em !important;
+        font-weight: bold !important;
+        border-radius: 8px !important;
+        border: 1px solid #ccc !important;
+    }
+    .rank-btn.secondary {
+        background: #f0f0f0 !important;
+        color: #333 !important;
+    }
+    .rank-btn.secondary:hover {
+        background: #e0e0e0 !important;
+        border-color: #bbb !important;
+    }
+    .absolute-eval-group {
+        border: 1px solid #ddd;
+        border-radius: 8px;
+        padding: 15px;
+        margin-top: 20px;
+    }
+    """
+
+    with gr.Blocks(title="Expression Evaluation Experiment", css=css) as app:
+        gr.Markdown("# Text-to-Expression Evaluation Experiment / テキストからの表情生成 評価実験")
+
+        with gr.Tabs() as tabs:
+            with gr.TabItem("1. Setup / セットアップ") as tab_setup:
+                gr.Markdown("## (A) Participant Information / 参加者情報")
+                gr.Markdown("Please enter your participant ID and click 'Confirm'. / 参加者IDを入力して「確定」を押してください。")
+                with gr.Row():
+                    participant_id_input = gr.Textbox(label="Participant ID", placeholder="e.g., P01")
+                    confirm_id_btn = gr.Button("Confirm / 確定", variant="primary")
+                setup_warning = gr.Markdown(visible=False)
+                with gr.Group(visible=False) as setup_main_group:
+                    gr.Markdown("---")
+                    gr.Markdown("## (B) Instructions & Data Loading / 注意事項とデータ読み込み")
+                    gr.Markdown(
+                        """<div style='padding: 15px; border: 1px solid #f0ad4e; border-radius: 5px; background-color: #fcf8e3;'>
+                        <h4>注意事項 / Instructions</h4>
+                        <ul>
+                            <li><b>この作業はPCで行ってください。/ Please perform this task on a PC.</b></li>
+                            <li>途中で止めずに最後まで続けてください。ファイルをアップロードして完了となります。/ Please continue until the end. The experiment is complete when you upload the file.</li>
+                            <li>ブラウザーをリロードしないでください (データが破損します)。/ Do not reload the browser (this will corrupt the data).</li>
+                        </ul></div>""")
+                    gr.Markdown(
+                        "Click the button below to load your evaluation data. / 下のボタンを押して、評価データを読み込んでください。")
+                    load_data_btn = gr.Button("Load Data / データ読み込み", variant="primary")
+                    setup_status = gr.Markdown("Waiting to start...")
+
+            with gr.TabItem("2. Evaluation / 評価", interactive=False) as tab_evaluation:
+                progress_text = gr.Markdown("Prompt 0 / 0")
+
+                image_components = []
+                rank_buttons = []
+                with gr.Row(equal_height=False):
+                    for label in IMAGE_LABELS:
+                        with gr.Column(scale=1):
+                            with gr.Group():
+                                gr.Markdown(f"<div class='image-label' style='text-align: center;'>{label}</div>")
+                                img = gr.Image(type="filepath", show_label=False, height=300)
+                                image_components.append(img)
+                                with gr.Row(elem_classes="rank-btn-row"):
+                                    for rank_val in range(1, 6):
+                                        btn = gr.Button(str(rank_val), variant='secondary', elem_classes="rank-btn")
+                                        rank_buttons.append(btn)
+
+                prompt_display = gr.Markdown("## \"Prompt Text Here\"", elem_classes="prompt-display")
+                gr.Markdown(
+                    "<p class='prompt-sub-guidance'>You may use AI or web search for the meaning of the text. However, please do not ask an AI about the emotion of the image itself.<br>意味についてはAIに聞いたりネット検索しても構いません。ただし、画像そのものの感情をAIに尋ねるのを止めてください。</p>")
+                guidance_display = gr.Markdown("### Guidance", elem_classes="rank-guidance")
+                error_display = gr.Markdown(visible=False)
+
+                gr.Markdown(
+                    """
+                    <b>ランキングの付け方 / How to Rank:</b>
+                    <ul>
+                        <li><b>全く同じ表情の画像には、同じ順位</b>を付けてください。(Assign the <b>same rank</b> to identical expressions.)</li>
+                        <li><b>少しでも違う表情の画像には、違う順位</b>を付けてください。(Assign <b>different ranks</b> to different expressions.)</li>
+                        <li><b>必ず1位から</b>順位を付けてください。(You <b>must</b> assign a rank of '1' to at least one image.)</li>
+                        <li>同順位がある場合、<b>その人数分だけ次の順位を飛ばしてください</b>。(When you have ties, <b>skip the next rank(s) accordingly</b>.)
+                            <ul>
+                                <li>例1: 1位が2つある場合、次は3位になります (Ex. 1: If there are two '1st' places, the next rank is '3rd'. e.g., <code>1, 1, 3, 4, 5</code>).</li>
+                                <li>例2: 1位が1つ、2位が3つある場合、次は5位になります (Ex. 2: If there is one '1st' and three '2nd' places, the next rank is '5th'. e.g., <code>1, 2, 2, 2, 5</code>).</li>
+                            </ul>
+                        </li>
+                    </ul>
+                    """,
+                    elem_classes="rank-instruction"
+                )
+
+                # ▼▼▼ 修正: 絶対評価（Best）のUI ▼▼▼
+                with gr.Group(visible=False, elem_classes="absolute-eval-group") as absolute_eval_group_best:
+                    gr.Markdown("---")
+                    gr.Markdown(
+                        "#### 絶対評価 (Best) / Absolute Score (Best)\n最もテキストと一致している画像について、どのていど一致しているかを評価してください。\n(Please evaluate the degree of alignment for the image that **best** matches the text.)")
+                    absolute_score_buttons = []
+                    with gr.Row():
+                        with gr.Column(scale=1):
+                            gr.Markdown(
+                                "<p style='text-align: right; margin-top: 10px;'>1 (全く一致してない / Not at all)</p>")
+                        with gr.Column(scale=3):
+                            with gr.Row(elem_classes="rank-btn-row"):
+                                for i in range(1, 8):
+                                    btn = gr.Button(str(i), variant='secondary', elem_classes="rank-btn")
+                                    absolute_score_buttons.append(btn)
+                        with gr.Column(scale=1):
+                            gr.Markdown("<p style='text-align: left; margin-top: 10px;'>7 (完全に一致 / Absolutely)</p>")
+
+                # ▼▼▼ 追加: 絶対評価（Worst）のUI ▼▼▼
+                with gr.Group(visible=False, elem_classes="absolute-eval-group") as absolute_eval_group_worst:
+                    gr.Markdown(
+                        "#### 絶対評価 (Worst) / Absolute Score (Worst)\n最もテキストと一致していない画像について、どのていど一致していないかを評価してください。\n(Please evaluate the degree of alignment for the image that **least** matches the text.)")
+                    absolute_score_worst_buttons = []
+                    with gr.Row():
+                        with gr.Column(scale=1):
+                            gr.Markdown(
+                                "<p style='text-align: right; margin-top: 10px;'>1 (全く一致してない / Not at all)</p>")
+                        with gr.Column(scale=3):
+                            with gr.Row(elem_classes="rank-btn-row"):
+                                for i in range(1, 8):
+                                    btn = gr.Button(str(i), variant='secondary', elem_classes="rank-btn")
+                                    absolute_score_worst_buttons.append(btn)
+                        with gr.Column(scale=1):
+                            gr.Markdown("<p style='text-align: left; margin-top: 10px;'>7 (完全に一致 / Absolutely)</p>")
+
+                with gr.Row():
+                    prev_btn = gr.Button("← Previous / 前へ", interactive=False)
+                    next_btn = gr.Button("Save & Next / 保存して次へ →", variant="primary")
+
+            with gr.TabItem("3. Export / エクスポート", interactive=False) as tab_export:
+                gr.Markdown("## (C) Final Comments & Export / 最終コメントとエクスポート")
+                gr.Markdown(
+                    "Thank you for completing the evaluation. Please provide the reasoning for your judgments for each criterion below. / 評価お疲れ様でした。以下の各評価基準について、判断の理由をご記入ください。")
+
+                with gr.Group():
+                    gr.Markdown("#### Reasoning for Judgments (Required) / 判断理由（必須）")
+                    alignment_reason_box = gr.Textbox(label="Alignment / 一致度", lines=3,
+                                                      placeholder="Why did you rank them this way for alignment? / なぜ一致度について、このような順位付けをしましたか？")
+                    naturalness_reason_box = gr.Textbox(label="Naturalness / 自然さ", lines=3,
+                                                        placeholder="Why did you rank them this way for naturalness? / なぜ自然さについて、このような順位付けをしましたか？")
+                    attractiveness_reason_box = gr.Textbox(label="Attractiveness / 魅力度", lines=3,
+                                                           placeholder="Why did you rank them this way for attractiveness? / なぜ魅力度について、このような順位付けをしましたか？")
+
+                with gr.Group():
+                    gr.Markdown("#### Overall Comments (Optional) / 全体的な感想（任意）")
+                    optional_comment_box = gr.Textbox(label="Any other comments? / その他、実験全体に関するご意見・ご感想",
+                                                      lines=4,
+                                                      placeholder="e.g., 'Image B often looked the most natural.' / 例：「Bの画像が最も自然に見えることが多かったです。」")
+
+                gr.Markdown("---")
+                gr.Markdown(
+                    "Finally, click the button below to export your results. / 最後に、下のボタンを押して結果をエクスポートしてください。")
+                export_btn = gr.Button("Export Results / 結果をエクスポート", variant="primary")
+                download_file = gr.File(label="Download JSON", visible=False)
+                export_status = gr.Markdown()
+
+        # --- Event Handlers ---
+        def check_and_confirm_id(pid):
+            pid = pid.strip()
+            if re.fullmatch(r"P\d{2}", pid):
+                GLOBAL_STATE["participant_id"] = pid
+                return gr.update(visible=False), gr.update(visible=True)
+            else:
+                error_msg = "<p class='feedback red'>Invalid ID. Must be 'P' followed by two digits (e.g., P01). / 無効なIDです。「P」と数字2桁の形式（例: P01）で入力してください。</p>"
+                return gr.update(value=error_msg, visible=True), gr.update(visible=False)
+
+        confirm_id_btn.click(check_and_confirm_id, [participant_id_input], [setup_warning, setup_main_group])
+        load_data_btn.click(load_evaluation_data, [participant_id_input], [setup_status, load_data_btn, tab_evaluation])
+
+        # ▼▼▼ 修正: all_eval_outputs に新しいUIコンポーネントを追加 ▼▼▼
+        all_eval_outputs = [
+            progress_text, prompt_display, guidance_display, error_display, *image_components,
+            *rank_buttons,
+            absolute_eval_group_best, *absolute_score_buttons,
+            absolute_eval_group_worst, *absolute_score_worst_buttons,
+            prev_btn, next_btn
+        ]
+
+        btn_idx = 0
+        for label in IMAGE_LABELS:
+            for rank_val in range(1, 6):
+                btn = rank_buttons[btn_idx]
+                btn.click(
+                    partial(handle_rank_button_click, label, rank_val),
+                    [],
+                    rank_buttons
+                )
+                btn_idx += 1
+
+        for i, btn in enumerate(absolute_score_buttons):
+            btn.click(
+                partial(handle_absolute_score_click, i + 1),
+                [],
+                absolute_score_buttons
+            )
+
+        # ▼▼▼ 追加: 新しいボタンのイベントハンドラを接続 ▼▼▼
+        for i, btn in enumerate(absolute_score_worst_buttons):
+            btn.click(
+                partial(handle_absolute_score_worst_click, i + 1),
+                [],
+                absolute_score_worst_buttons
+            )
+
+        tab_evaluation.select(display_current_prompt_and_criterion, [], all_eval_outputs)
+
+        # ▼▼▼ 修正: next_btn の出力に tab_export を追加 ▼▼▼
+        next_btn.click(validate_and_navigate, [], [tab_export, *all_eval_outputs])
+
+        prev_btn.click(navigate_previous, [], all_eval_outputs)
+
+        export_tab_interactive_components = [alignment_reason_box, naturalness_reason_box, attractiveness_reason_box,
+                                             optional_comment_box, export_btn]
+
+        def on_select_export_tab():
+            # end_time is set only when all evaluations are complete
+            if GLOBAL_STATE.get("end_time"):
+                return [gr.update(interactive=True)] * 5
+            # This logic is now handled by next_btn click, but kept as a fallback.
+            return [gr.update(interactive=False)] * 5
+
+        tab_export.select(on_select_export_tab, [], export_tab_interactive_components)
+
+        export_btn.click(
+            export_results,
+            [participant_id_input, alignment_reason_box, naturalness_reason_box, attractiveness_reason_box,
+             optional_comment_box],
+            [download_file, export_status]
+        )
+
+    return app
+
+if __name__ == "__main__":
+    os.makedirs(LOG_DIR, exist_ok=True)
+    log_file_path = os.path.join(LOG_DIR, "evaluation_ui_log_{time}.log")
+
+    random.seed(datetime.now().timestamp())
+    logger.remove()
+    logger.add(sys.stderr, level="INFO")
+    logger.add(log_file_path, rotation="10 MB")
+
+    app = create_gradio_interface()
+    app.launch(share=True, debug=True)

combined_data/blendshape_annotation_preprocess.py

@@ -0,0 +1,114 @@
+import os
+import json
+import torch
+import numpy as np
+from PIL import Image
+from tqdm import tqdm
+from loguru import logger
+from typing import Dict, List, Tuple
+
+# ---- あなたの定義済み Dataset, transforms があるなら import ----
+# from your_dataset import BlendShapeDataset, image_transform
+
+def unify_index(idx: int, group_size: int) -> int:
+    """
+    BlendShape の選択 index が group_size と同じ場合、-1 (none) に変換して返す。
+    """
+    if idx == group_size:
+        return -1
+    return idx
+
+def build_combination_to_filename(
+    teacher_data_file: str,
+    meta_file: str,
+) -> Tuple[Dict[Tuple[int, ...], str], List[int]]:
+    """
+    BlendShapeData.json を読み込み、
+    例:
+       { (g0_idx, g1_idx, g2_idx): "000001.png", ... }
+    のような辞書を作成して返す。
+
+    さらに各グループのサイズ (blendShapeNames数 + 1) のリスト group_sizes も返す。
+    """
+    with open(meta_file, "r", encoding="utf-8") as f:
+        meta = json.load(f)
+    blend_shape_groups = meta["blendShapeGroupsMeta"]
+    # group_sizes[i] = len(そのグループの blendShapeNames) + 1(none枠)
+    group_sizes = [len(g["blendShapeNames"]) + 1 for g in blend_shape_groups]
+
+    # teacher_data_file 読み込み
+    with open(teacher_data_file, "r", encoding="utf-8") as f:
+        teacher_data = json.load(f)
+    data_list = teacher_data["dataList"]
+
+    combination_to_filename = {}
+    for data in data_list:
+        photo_filename = data["photoFileName"]
+        blendShapeSelections = data["blendShapeSelectionsPerGroup"]
+
+        # グループ順に selectedBlendShapeIndex を取得しつつ、-1の場合は-1のまま、
+        # group_sizeと一致していたら-1へ変換(理想的には -1 しか登場しない想定だが一応対応)
+        combo = []
+        for group_idx, selection in enumerate(blendShapeSelections):
+            sel_idx = selection["selectedBlendShapeIndex"]
+            # group_sizes[group_idx] と同じなら none として -1
+            sel_idx = unify_index(sel_idx, group_sizes[group_idx])
+            combo.append(sel_idx)
+
+        combo = tuple(combo)  # dictのキーにするのでtuple化
+        combination_to_filename[combo] = photo_filename
+
+    return combination_to_filename, group_sizes
+
+
+
+def main_offline_precompute():
+    """
+    1. Dataset(JSON)から (組み合わせ -> filename) を作る
+    2. CLIP モデルロード
+    3. filename -> embedding
+    4. ペアワイズ類似度
+    5. 保存
+    """
+    import argparse
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--image_dir", type=str, default="lapwing/images")
+    parser.add_argument("--meta_file", type=str, default="lapwing/texts/BlendShapeGroupsMeta.json")
+    parser.add_argument("--teacher_data_file", type=str, default="lapwing/texts/BlendShapeData.json")
+    parser.add_argument("--clip_model_name", type=str, default="ViT-L-14")
+    parser.add_argument("--clip_pretrained", type=str, default="openai")
+    parser.add_argument("--batch_size", type=int, default=8)
+    parser.add_argument("--device", type=str, default="cuda")
+    parser.add_argument("--out_comb2fn", type=str, default="combination_to_filename.json")
+    parser.add_argument("--out_sims", type=str, default="pairwise_clip_sims.json")
+
+    args = parser.parse_args()
+
+    # 1. 組み合わせ->filename辞書の作成
+    combination_to_filename, group_sizes = build_combination_to_filename(
+        teacher_data_file=args.teacher_data_file,
+        meta_file=args.meta_file,
+    )
+
+
+    # 5. 保存 (JSON形式)
+    #   5.1 (combination -> filename)
+    #       group_sizes も保存しておくと後段のオンライン時に参照しやすい
+    #       tupleは文字列化する必要あり
+    comb2fn_dict = {
+        "group_sizes": group_sizes,
+        "mapping": {
+            ",".join(map(str, comb)): fn
+            for comb, fn in combination_to_filename.items()
+        }
+    }
+    with open(args.out_comb2fn, "w", encoding="utf-8") as f:
+        json.dump(comb2fn_dict, f, ensure_ascii=False, indent=2)
+
+    # データ出力
+    for combo, filename in combination_to_filename.items():
+        logger.info(f"Combination: {combo}, Filename: {filename}")
+
+
+if __name__ == "__main__":
+    main_offline_precompute()

combined_data/combination_to_filename.json

@@ -0,0 +1,80 @@
+{
+  "group_sizes": [
+    8,
+    9
+  ],
+  "mapping": {
+    "-1,-1": "000001.png",
+    "-1,0": "000002.png",
+    "-1,1": "000003.png",
+    "-1,2": "000004.png",
+    "-1,3": "000005.png",
+    "-1,4": "000006.png",
+    "-1,5": "000007.png",
+    "-1,6": "000008.png",
+    "-1,7": "000009.png",
+    "0,-1": "000010.png",
+    "0,0": "000011.png",
+    "0,1": "000012.png",
+    "0,2": "000013.png",
+    "0,3": "000014.png",
+    "0,4": "000015.png",
+    "0,5": "000016.png",
+    "0,6": "000017.png",
+    "0,7": "000018.png",
+    "1,-1": "000019.png",
+    "1,0": "000020.png",
+    "1,1": "000021.png",
+    "1,2": "000022.png",
+    "1,3": "000023.png",
+    "1,4": "000024.png",
+    "1,5": "000025.png",
+    "1,6": "000026.png",
+    "1,7": "000027.png",
+    "2,-1": "000028.png",
+    "2,0": "000029.png",
+    "2,1": "000030.png",
+    "2,2": "000031.png",
+    "2,3": "000032.png",
+    "2,4": "000033.png",
+    "2,5": "000034.png",
+    "2,6": "000035.png",
+    "2,7": "000036.png",
+    "3,-1": "000037.png",
+    "3,0": "000038.png",
+    "3,1": "000039.png",
+    "3,2": "000040.png",
+    "3,3": "000041.png",
+    "3,4": "000042.png",
+    "3,5": "000043.png",
+    "3,6": "000044.png",
+    "3,7": "000045.png",
+    "4,-1": "000046.png",
+    "4,0": "000047.png",
+    "4,1": "000048.png",
+    "4,2": "000049.png",
+    "4,3": "000050.png",
+    "4,4": "000051.png",
+    "4,5": "000052.png",
+    "4,6": "000053.png",
+    "4,7": "000054.png",
+    "5,-1": "000055.png",
+    "5,0": "000056.png",
+    "5,1": "000057.png",
+    "5,2": "000058.png",
+    "5,3": "000059.png",
+    "5,4": "000060.png",
+    "5,5": "000061.png",
+    "5,6": "000062.png",
+    "5,7": "000063.png",
+    "6,-1": "000064.png",
+    "6,0": "000065.png",
+    "6,1": "000066.png",
+    "6,2": "000067.png",
+    "6,3": "000068.png",
+    "6,4": "000069.png",
+    "6,5": "000070.png",
+    "6,6": "000071.png",
+    "6,7": "000072.png"
+  }
+}