import os
import json
import glob
from pathlib import Path
from collections import defaultdict

# ================= 配置区域 =================
DATA_ROOT = "/opt/liblibai-models/user-workspace2/users/dxy/robotic/data" 
REF_IMG_ROOT = "/opt/liblibai-models/user-workspace2/users/dxy/robotic/robot_ref_img"

# 修改点 1: 不再需要单一的 OUTPUT_JSONL，而是指定一个输出目录
OUTPUT_DIR = "./json_outputs" 

VIDEO_EXTS = {".mp4", ".mov", ".avi", ".mkv"}
IMAGE_EXTS = {".jpg", ".jpeg", ".png", ".webp"}
# ===========================================

def generate_strict_paired_metadata():
    # 确保输出目录存在
    os.makedirs(OUTPUT_DIR, exist_ok=True)

    # -------------------------------------------------
    # 1. 建立参考图索引 (代码保持不变)
    # -------------------------------------------------
    ref_images_map = {}
    if os.path.exists(REF_IMG_ROOT):
        print(f"Scanning reference images in {REF_IMG_ROOT}...")
        for img_file in os.listdir(REF_IMG_ROOT):
            name, ext = os.path.splitext(img_file)
            if ext.lower() in IMAGE_EXTS:
                ref_images_map[name] = os.path.abspath(os.path.join(REF_IMG_ROOT, img_file))
    else:
        print(f"Warning: Reference image directory '{REF_IMG_ROOT}' not found.")

    # -------------------------------------------------
    # 2. 建立严格的上下文视频索引 (代码保持不变)
    # -------------------------------------------------
    context_map = defaultdict(list)
    print(f"\nScanning videos in {DATA_ROOT}...")
    
    for root, dirs, files in os.walk(DATA_ROOT):
        videos = [f for f in files if os.path.splitext(f)[1].lower() in VIDEO_EXTS]
        if not videos:
            continue
            
        rel_path = os.path.relpath(root, DATA_ROOT)
        path_parts = Path(rel_path).parts
        
        if len(path_parts) < 4:
            continue
            
        scene = path_parts[0]
        location = path_parts[1]
        camera = path_parts[2]
        # 你的逻辑：取最后一部分文件夹名为角色名
        character = path_parts[-1] 

        for vid in videos:
            motion_name = os.path.splitext(vid)[0]
            full_path = os.path.join(root, vid)
            unique_key = (scene, location, camera, motion_name)
            
            context_map[unique_key].append({
                "char": character,
                "path": full_path,
                "motion": motion_name
            })

    print(f"Found {len(context_map)} unique action contexts.")
    
    # -------------------------------------------------
    # 3. 生成全排列配对 (修改点：使用字典按角色分类)
    # -------------------------------------------------
    # key: 角色名 (例如 'unitree-g1'), value: 该角色的数据列表
    results_by_char = defaultdict(list)
    
    for key, video_list in context_map.items():
        scene, loc, cam, motion = key
        
        if len(video_list) < 2:
            continue
            
        # 这里的 target_item 就是“目标视频”
        for target_item in video_list:
            target_char = target_item["char"]
            target_path = target_item["path"]
            
            if target_char not in ref_images_map:
                continue
            ref_img_path = ref_images_map[target_char]
            
            # 寻找 Control 视频 (其他角色做的同一个动作)
            for source_item in video_list:
                source_char = source_item["char"]
                
                # 排除自己 (只用别的外观做 Control)
                if source_char == target_char:
                    continue
                    
                control_path = source_item["path"]
                
                # Prompt 逻辑保持不变
                prompt_path = os.path.splitext(target_path)[0] + ".txt"
                prompt = ""
                if os.path.exists(prompt_path):
                    try:
                        with open(prompt_path, 'r', encoding='utf-8') as f:
                            prompt = f.read().strip()
                    except:
                        pass
                
                if not prompt:
                    safe_motion = motion.replace("_", " ")
                    prompt = f"A robot {target_char} is {safe_motion} in {scene} at {loc}"

                entry = {
                    "video": target_path,
                    "control_video": control_path,
                    "prompt": prompt,
                    "appearance_id": target_char,
                    "ref_image_path": ref_img_path,
                }
                
                # 【修改核心】：将数据存入对应 target_char 的列表中
                results_by_char[target_char].append(entry)

    # -------------------------------------------------
    # 4. 分文件保存 (修改点：循环写入多个文件)
    # -------------------------------------------------
    print(f"\nGeneration complete! Writing files to {OUTPUT_DIR}...")
    
    total_files = 0
    for char_name, entries in results_by_char.items():
        if not entries:
            continue
            
        # 构造文件名，例如: data_unitree-g1.jsonl
        # 使用 replace 处理一下文件名里可能不合法的字符
        safe_char_name = char_name.replace("/", "_").replace("\\", "_")
        filename = f"data_{safe_char_name}.jsonl"
        file_path = os.path.join(OUTPUT_DIR, filename)
        
        with open(file_path, 'w', encoding='utf-8') as f:
            for item in entries:
                f.write(json.dumps(item, ensure_ascii=False) + "\n")
        
        print(f" -> Saved {filename} ({len(entries)} samples)")
        total_files += 1
    
    print(f"\nAll done! Created {total_files} JSONL files.")

if __name__ == "__main__":
    generate_strict_paired_metadata()