upload

make_artist_all_in_one_dataset_from_hf.py

@@ -0,0 +1,303 @@
+import os
+import argparse
+import shutil
+import threading
+import logging
+from pathlib import Path
+from huggingface_hub import HfApi, hf_hub_download
+from tqdm import tqdm
+from queue import Queue
+
+
+def get_logging(level = None) -> logging.Logger:
+    logger = logging.getLogger('DB-Maker')
+    if level is None:
+        level = logging.DEBUG
+    logger.setLevel(level)  # 设置日志级别
+
+    # 创建一个控制台处理器
+    console_handler = logging.StreamHandler()
+
+    # 创建一个格式器，并定义日志输出格式
+    formatter = logging.Formatter('%(asctime)s - %(name)s - %(levelname)s - %(message)s')
+    console_handler.setFormatter(formatter)
+
+    # 将处理器添加到记录器
+    logger.addHandler(console_handler)
+    return logger
+
+
+def get_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser()
+    normalized_filepath = lambda filepath: str(Path(filepath).absolute().as_posix())
+
+    parser.add_argument("--hf-token", type = str, default = None, help = "huggingFace Token")
+    parser.add_argument("--repo", type = str, default = None, help = "仓库的名称")
+    parser.add_argument("--repo-type", type = str, default = None, help = "HuggingFace 仓库的种类, 如: model, dataset")
+    parser.add_argument("--local-path", type = normalized_filepath, default = None, help = "本地要上传文件的路径")
+    parser.add_argument("--folder", type = str, default = None, help = "指定下载仓库的某个文件夹内容")
+
+    return parser.parse_args()
+
+
+def get_hf_file_list(repo, repo_type, hf_token) -> list:
+    logger.info(f"获取 {repo} 中所有的文件列表中")
+    api = HfApi()
+    model_list = api.list_repo_files(
+        repo_id = repo,
+        repo_type = repo_type,
+        token = hf_token
+    )
+    logger.info(f"{repo} 中所有文件的数量: {len(model_list)}")
+    return model_list
+
+
+def download_hf_file(repo_id: str, repo_type: str, filename: str, local_dir: str, hf_token: str) -> None:
+    try:
+        logger.info(f"下载 {filename} 到 {local_dir} 中")
+        hf_hub_download(
+            repo_id = repo_id,
+            repo_type = repo_type,
+            filename = filename, 
+            local_dir_use_symlinks = False, 
+            local_dir = local_dir,
+            token = hf_token
+        )
+        logger.info(f"{filename} 下载成功")
+    except:
+        logger.error(f"{filename} 下载失败")
+
+
+def filter_file(file_list: list, start_dir: str) -> list:
+    filter_file_list = []
+    for f in file_list:
+        if f.startswith(start_dir):
+            filter_file_list.append(f)
+
+    logger.info(f"{start_dir} 中的文件数量: {len(filter_file_list)}")
+    return filter_file_list
+
+
+def insert_tag(path: str, tag:str) -> None:
+    content = []
+    if not os.path.exists(path):
+        logger.info(f"{path} 文件未找到")
+        return None
+
+    try:
+        with open(path, "r", encoding="utf8") as f:
+            for i in f.readlines():
+                content.append(i)
+
+        if ''.join(content).startswith(tag):
+            return
+
+        with open(path, "w", encoding="utf8") as f:
+            f.write(f"{tag}, {''.join(content)}")
+    except Exception as e:
+        logger.error(f"打开 {path} 失败", e)
+
+
+def get_all_file(directory: str) -> list:
+    file_list = []
+    for dirname, _, filenames in os.walk(directory):
+        for filename in filenames:
+            file_list.append(os.path.join(dirname, filename))
+    return file_list
+
+
+def process_dataset_tag(dataset_path: str, tag: str) -> None:
+    if not os.path.exists(dataset_path):
+        print(f"{dataset_path} 未找到")
+        return
+
+    file_list = get_all_file(dataset_path)
+
+    # 过滤文件
+    tag_file = []
+    for file in file_list:
+        if file.endswith(".txt"):
+            tag_file.append(file)
+
+    for file in tqdm(tag_file, desc=f"添加 {tag} 标签"):
+        insert_tag(file, tag)
+
+
+def make_dataset(repo: str, repo_type: str, hf_token: str, hf_repo_file_list: list, hf_folder: str, path:str, repeat: int, tag: str) -> None:
+    dataset_path = os.path.join(path, hf_folder)
+
+    if os.path.exists(os.path.join(os.path.dirname(dataset_path), f"{str(repeat)}_{hf_folder}")):
+        logger.info(f"{hf_folder} 已存在")
+        return
+
+    logger.info(f"处理数据集: {hf_folder}")
+
+    hf_list = filter_file(hf_repo_file_list, hf_folder)
+    task_list = []
+    for filename in tqdm(hf_list, desc = "创建下载任务"):
+        task_list.append([repo, repo_type, filename, path, hf_token])
+    model_downloader = ModelDownload(task_list)
+    model_downloader.start_threads()
+
+    process_dataset_tag(dataset_path, tag)
+    shutil.move(dataset_path, os.path.join(os.path.dirname(dataset_path), f"{str(repeat)}_{hf_folder}"))
+
+
+class ModelDownload:
+    def __init__(self, urls) -> None:
+        self.urls = urls
+        self.queue = Queue()
+        self.total_urls = len(urls)  # 记录总的URL数
+        self.downloaded_count = 0  # 记录已下载的数量
+        self.lock = threading.Lock()  # 创建锁以保护对下载计数器的访问
+        self.progress_bar = tqdm(total=self.total_urls, desc="下载进度", unit="file")  # 创建进度条实例
+
+
+    def worker(self):
+        while True:
+            url = self.queue.get()
+            if url is None:
+                break
+            download_hf_file(url[0], url[1], url[2], url[3], url[4])
+            self.queue.task_done()
+            with self.lock:  # 访问共享资源时加锁
+                self.downloaded_count += 1
+                self.print_progress()  # 打印进度
+                # self.progress_bar.update(1)  # 更新进度条
+
+
+    def print_progress(self):
+        progress = (self.downloaded_count / self.total_urls) * 100
+        logger.info(f"下载进度: {self.downloaded_count}/{self.total_urls} ({progress:.2f}%)")
+
+
+    def finish(self):
+        self.progress_bar.close()  # 关闭进度条
+
+
+    def start_threads(self, num_threads=16):
+        threads = []
+        for _ in range(num_threads):
+            thread = threading.Thread(target=self.worker)
+            thread.start()
+            threads.append(thread)
+
+        for url in self.urls:
+            self.queue.put(url)
+
+        self.queue.join()
+
+        for _ in range(num_threads):
+            self.queue.put(None)
+
+        for thread in threads:
+            thread.join()
+
+        self.finish()  # 完成后关闭进度条
+
+
+def main():
+    global logger
+    logger = get_logging()
+
+    repo = "licyk/image_training_set"
+    repo_type = "dataset"
+    hf_token = None
+    dataset_path = os.path.join(os.path.dirname(os.path.abspath(__file__)), "artist_all_in_one")
+    dataset_list = [
+        ["gan_cheng_1", "nachoneko", 2], # 125
+        ["rurudo_1", "rurudo", 7], # 40
+        ["deyui_1", "deyui", 2], # 148
+        ["tyomimas_1", "tyomimas", 1], # 238
+        ["linhai_1", "linhai", 7], # 40
+        ["viclim-monou_1", "viclim monou", 4], # 74
+        ["yeurei_1", "yeurei", 4], # 69
+        ["asagi_0398_1", "asagi0398", 6], # 44
+        ["Noyu_1", "noyu", 3], # 86
+        ["gomalio_y_1", "gomalio y", 4], # 74
+        ["qian_qian_jie_1", "qianqianjie", 4], # 64
+        ["sheya_1", "sheya", 3], # 85
+        ["Blue_Archive_Official_2", "blue archive official", 1], # 262, max
+        ["sei_umehara_1", "sei umehara", 2], # 132
+        ["rafa_1", "rafa", 5], # 52
+        ["yume335_1", "yume335", 6], # 43
+        ["Yunouou10_1", "yunouou10", 11], # 23
+        ["Renren_1", "renren", 6], # 47
+        ["AL17460600_1", "al17460600", 5], # 49
+        ["fruitsrabbit_1", "fruitsrabbit", 2], # 158
+        ["void_0_1", "void0", 1], # 234
+        ["ekureea_1", "ekureea", 4], # 74
+        ["nocopyrightgirl_1", "nocopyrightgirl", 2], # 121
+        ["KonYa666_1", "konya666", 3], # 109
+        ["momoimon_1", "momoimon", 3], # 89
+        ["a20_1", "a20", 4], # 64
+        ["ikari_1", "ikari", 2], # 123
+        ["ningen_mame_1", "ningen mame", 2], # 139
+        ["jyt_1", "jyt", 1], # 195
+        ["suzume_1", "suzume", 1], # 197
+        ["deieaf_1", "deieaf", 2], # 168
+        ["hk3_1", "hk3", 3], # 80
+        ["matchach_1", "matchach", 1], # 218
+        ["quan_1", "quan", 2], # 126
+        ["fuzichoco_1", "fuzichoco", 2], # 133
+        ["torino_1", "torino", 2], # 186
+        ["negimapurinn_1", "negimapurinn", 3], # 94
+        ["anon_1", "anon", 3], # 103
+        ["honkai_star_rail_expression_1", "honkai star rail expression", 2], # 167
+        ["lanubis_2", "lanubis", 5], # 56
+        ["wanke_1", "wanke", 3], # 96
+        ["ixy_1", "ixy", 2], # 167
+        ["jirujiaru826_1", "jirujiaru826", 3], # 94
+        ["Sco_ttie_1", "sco ttie", 2], # 183
+        ["lplpsg_1", "lplpsg", 6], # 48
+        ["sakurapion_1", "sakurapion", 4], # 67
+        # ["pvc_1", "pvc", 1], # 546
+        ["guo_jiang_mullay_1", "guo jiang mullay", 7], # 40
+        # ["wlop_1", "wlop", 1], # 210
+        ["Colon_BR_1", "colon br", 3], # 90
+        ["DaylightAllure_1", "daylightallure", 3], # 109
+        ["kagawayusaku_1", "kagawayusaku", 3], # 105
+        ["ogipote_1", "ogipote", 4], # 60
+        ["shiratama_shiratamaco_1", "shiratama shiratamaco", 2], # 132
+        ["yoneyama_mai_1", "yoneyama mai", 4], # 60
+        ["genshin_cg_flat_color_style_1", "genshin cg flat color style", 1], # 219
+        ["Batory_ken_1", "batory ken", 5], # 52
+        # ["Nakkar7_1", "nakkar7", 14], # 40
+        ["siro90414_1", "siro90414", 2], # 161
+        ["sunfish_1", "sunfish", 4], # 64
+        ["Anmi_1", "anmi",5], # 59
+        ["hakua_aa_1", "hakua aa", 3], # 81
+        ["WERI_1", "weri", 5], # 56
+    ]
+    
+
+    logger.info(f"仓库: {repo}")
+    logger.info(f"仓库种类: {repo_type}")
+    logger.info(f"训练集目录: {dataset_path}")
+
+    hf_repo_file_list = get_hf_file_list(repo, repo_type, hf_token)
+
+    for dataset in tqdm(dataset_list, desc="处理训练集"):
+        make_dataset(
+            repo=repo,
+            repo_type=repo_type,
+            hf_token=hf_token,
+            hf_repo_file_list=hf_repo_file_list,
+            hf_folder=dataset[0],
+            path=dataset_path,
+            repeat=dataset[2],
+            tag=dataset[1],
+        )
+    logger.info(f"训练集处理完成, 路径: {dataset_path}")
+    # hf_list = filter_file(hf_repo_file_list, folder)
+    # logging.info(f"下载 {repo} 中的文件中")
+    # task_list = []
+    # for filename in tqdm(hf_list, desc = "创建下载任务"):
+    #     task_list.append([repo, repo_type, filename, root_path, hf_token])
+
+    # model_downloader = ModelDownload(task_list)
+    # model_downloader.start_threads(num_threads=5)
+    # print(f"下载 {repo} 完成")
+
+
+main()