hjxdl 0.3.46__py3-none-any.whl → 0.3.48__py3-none-any.whl

hdl/_version.py

@@ -17,5 +17,5 @@ __version__: str
 __version_tuple__: VERSION_TUPLE
 version_tuple: VERSION_TUPLE
 
-__version__ = version = '0.3.46'
-__version_tuple__ = version_tuple = (0, 3, 46)
+__version__ = version = '0.3.48'
+__version_tuple__ = version_tuple = (0, 3, 48)

hdl/utils/vis_tools/scene_detect.py

@@ -0,0 +1,198 @@
+import os
+import json
+import base64
+from pathlib import Path
+# from PIL import Image
+# import openai
+# from scenedetect import VideoManager, SceneManager
+# from scenedetect.detectors import ContentDetector
+import cv2
+from tqdm import tqdm
+import pandas as pd
+import subprocess
+
+
+def detect_scenes_cli(video_path, output_dir, threshold=20):
+    video_path = Path(video_path)
+    output_dir = Path(output_dir)
+    # output_dir = video_path.parent
+    base_name = video_path.stem
+    csv_path = output_dir / f"{base_name}-Scenes.csv"
+
+    subprocess.run([
+        "scenedetect",
+        "-i", str(video_path),
+        "detect-content",
+        f"--threshold={threshold}",
+        "list-scenes",
+        "-o", str(output_dir)
+    ], check=True)
+
+    return csv_path
+
+
+def read_start_frames_from_csv(csv_path):
+    df = pd.read_csv(csv_path, skiprows=1)
+    return df['Start Frame'].astype(int).tolist()
+
+
+def extract_frames_with_cv(video_path, frame_numbers, output_dir, grid_size=(3, 3)):
+    Path(output_dir).mkdir(exist_ok=True)
+    cap = cv2.VideoCapture(video_path)
+    total_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
+
+    grid_w, grid_h = grid_size
+    num_required = grid_w * grid_h
+
+    for i, start_frame in tqdm(enumerate(frame_numbers), total=len(frame_numbers), desc="抽取关键帧"):
+        end_frame = frame_numbers[i + 1] if i < len(frame_numbers) - 1 else total_frames - 1
+
+        available_range = end_frame - start_frame
+        if available_range <= num_required + 2:
+            print(f"⚠️ 场景 {i} 太短（帧数不足 {num_required + 2}），跳过")
+            continue
+
+        # 从 start+1 到 end-1 中抽取 num_required 个等间隔帧
+        step = (end_frame - start_frame - 2) / (num_required - 1)
+        selected_frames = [int(start_frame + 1 + round(j * step)) for j in range(num_required)]
+
+        frames = []
+        for frame_num in selected_frames:
+            cap.set(cv2.CAP_PROP_POS_FRAMES, frame_num)
+            ret, frame = cap.read()
+            if not ret:
+                print(f"⚠️ 无法读取帧 {frame_num}，跳过场景 {i}")
+                frames = []
+                break
+            frames.append(frame)
+
+        if len(frames) != num_required:
+            print(f"⚠️ 场景 {i} 抽帧失败，不完整，跳过")
+            continue
+
+        # 拼成 grid_w * grid_h 图像
+        try:
+            rows = [cv2.hconcat(frames[y * grid_w:(y + 1) * grid_w]) for y in range(grid_h)]
+            grid_image = cv2.vconcat(rows)
+            output_img = f"{output_dir}/scene_{i:03d}.jpg"
+            cv2.imwrite(output_img, grid_image)
+        except Exception as e:
+            print(f"❌ 拼图失败 场景 {i}：{e}")
+
+    cap.release()
+
+
+def generate_json_template(frame_numbers, output_dir, output_path):
+    output = []
+    for i, frame in enumerate(frame_numbers):
+        output.append({
+            "start_frame": frame,
+            "image": f"scene_{i:03d}.jpg",
+            "description": "请描述这张图代表的场景"
+        })
+    with open(output_path, "w", encoding="utf-8") as f:
+        json.dump(output, f, indent=4, ensure_ascii=False)
+
+
+def describe_image(
+    image_path,
+    client,
+    model: str = "default_model",
+    sys_info: str = (
+        "你是一个擅长视频场景理解的 AI，请根据提供的拼图图像，生成统一格式的视频场景描述。\n"
+        "请包括：\n"
+        "1. 场景中发生的主要事件或动作。\n"
+        "2. 出现的人物（如有）及其行为。\n"
+        "3. 场景的背景和气氛。\n"
+        "请用简洁、客观、第三人称的方式描述，不要加入主观感受。\n"
+        "输出格式如下：\n"
+        "场景描述：XXX。"
+    )
+):
+    try:
+        with open(image_path, "rb") as f:
+            img_bytes = f.read()
+            b64_img = base64.b64encode(img_bytes).decode("utf-8")
+
+        response = client.chat.completions.create(
+            model=model,
+            messages=[
+                {"role": "system", "content": sys_info},
+                {
+                    "role": "user",
+                    "content": [
+                        {"type": "text", "text": "请根据图像生成一段客观的视频场景描述，内容包括人物、动作、背景、氛围。输出请以“场景描述：”开头。"},
+                        {"type": "image_url", "image_url": {"url": f"data:image/jpeg;base64,{b64_img}"}}
+                    ]
+                }
+            ],
+            max_tokens=200
+        )
+        return response.choices[0].message.content.strip()
+    except Exception as e:
+        print(f"❌ 处理图片 {image_path} 时出错: {e}")
+        return "描述生成失败"
+
+
+def fill_descriptions(
+    client,
+    input_json,
+    output_json,
+    output_dir
+):
+    with open(input_json, "r", encoding="utf-8") as f:
+        scenes = json.load(f)
+
+    for scene in tqdm(scenes, desc="生成场景描述"):
+        img_path = os.path.join(output_dir, scene["image"])
+        scene["description"] = describe_image(img_path, client=client)
+        scene.pop("image", None)
+
+    with open(output_json, "w", encoding="utf-8") as f:
+        json.dump(scenes, f, indent=4, ensure_ascii=False)
+
+    print(f"✅ 完成：{output_json}")
+
+
+class SceneDetector(object):
+    def __init__(
+        self,
+        client,
+        video_file,
+        pre_processing: bool = False,
+        temp_json: str = None,
+        final_json: str = None
+    ):
+        self.client = client
+        self.video_file = video_file
+        self.pre_processsing = pre_processing
+        self.temp_json = temp_json
+        self.final_json = final_json
+
+        if not self.temp_json:
+            self.temp_json = self.video_file + ".tmp.json"
+        if not self.final_json
+            self.final_json = self.video_file + ".final.json"
+
+        if self.pre_processsing:
+            self.pre_process()
+
+    def pre_process(self):
+        pass
+
+    def detect(
+        self,
+        out_dir
+    ):
+        output_csv = detect_scenes_cli(self.video_file, out_dir)
+        # df = pd.read_csv(output_csv, skiprows=1)
+        # df = read_start_frames_from_csv(output_csv)
+        starts = read_start_frames_from_csv(output_csv)
+        extract_frames_with_cv(self.video_file, starts, out_dir)
+        generate_json_template(starts, out_dir, self.temp_json)
+        fill_descriptions(
+            self.temp_json,
+            self.final_json,
+            out_dir
+        )
+

{hjxdl-0.3.46.dist-info → hjxdl-0.3.48.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: hjxdl
-Version: 0.3.46
+Version: 0.3.48
 Summary: A collection of functions for Jupyter notebooks
 Home-page: https://github.com/huluxiaohuowa/hdl
 Author: Jianxing Hu

{hjxdl-0.3.46.dist-info → hjxdl-0.3.48.dist-info}/RECORD

@@ -1,5 +1,5 @@
 hdl/__init__.py,sha256=GffnD0jLJdhkd-vo989v40N90sQbofkayRBwxc6TVhQ,72
-hdl/_version.py,sha256=MipZXgcW4nH_pXJSDrgcZpvolhZYMhJ7kQyobJyAtjA,513
+hdl/_version.py,sha256=HakMWBDNRn2Lw-dDW5Jqd_KgJ7xBCMwyZo_L2WGE1yM,513
 hdl/args/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 hdl/args/loss_args.py,sha256=s7YzSdd7IjD24rZvvOrxLLFqMZQb9YylxKeyelSdrTk,70
 hdl/controllers/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -142,10 +142,12 @@ hdl/utils/llm/vis.py,sha256=KuypafAuU4MEzJ5ywvopI0RcYjnyF_WKM_0oDUKbr8A,28808
 hdl/utils/llm/visrag.py,sha256=0i-VrxqgiV-J7R3VPshu9oc7-rKjFJOldYik3HDXj6M,10176
 hdl/utils/schedulers/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 hdl/utils/schedulers/norm_lr.py,sha256=bDwCmdEK-WkgxQMFBiMuchv8Mm7C0-GZJ6usm-PQk14,4461
+hdl/utils/vis_tools/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+hdl/utils/vis_tools/scene_detect.py,sha256=T9mQbxbZoeZZIqm8scXSg9-jvtKPzqyiVNtc_ykhxZE,6471
 hdl/utils/weather/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 hdl/utils/weather/weather.py,sha256=k11o6wM15kF8b9NMlEfrg68ak-SfSYLN3nOOflFUv-I,4381
-hjxdl-0.3.46.dist-info/licenses/LICENSE,sha256=lkMiSbeZHBQLB9LJEkS9-L3Z-LBC4yGnKrzHSG8RkPM,2599
-hjxdl-0.3.46.dist-info/METADATA,sha256=rJxecUIQvBpir518gAteLCMHTV5SjGMsvu29a8w03ao,1332
-hjxdl-0.3.46.dist-info/WHEEL,sha256=Nw36Djuh_5VDukK0H78QzOX-_FQEo6V37m3nkm96gtU,91
-hjxdl-0.3.46.dist-info/top_level.txt,sha256=-kxwTM5JPhylp06z3zAVO3w6_h7wtBfBo2zgM6YZoTk,4
-hjxdl-0.3.46.dist-info/RECORD,,
+hjxdl-0.3.48.dist-info/licenses/LICENSE,sha256=lkMiSbeZHBQLB9LJEkS9-L3Z-LBC4yGnKrzHSG8RkPM,2599
+hjxdl-0.3.48.dist-info/METADATA,sha256=s9uYj4yBltF7eBP1yy3kgcyJmDUEB2Nm-CQSzWsqkBY,1332
+hjxdl-0.3.48.dist-info/WHEEL,sha256=Nw36Djuh_5VDukK0H78QzOX-_FQEo6V37m3nkm96gtU,91
+hjxdl-0.3.48.dist-info/top_level.txt,sha256=-kxwTM5JPhylp06z3zAVO3w6_h7wtBfBo2zgM6YZoTk,4
+hjxdl-0.3.48.dist-info/RECORD,,