Upload 2 files

noinstance_copy_script.py

@@ -0,0 +1,63 @@
+'''
+
+'''
+
+import os
+import cv2
+import argparse
+import shutil
+from pathlib import Path
+from tqdm import tqdm
+
+from animeinsseg import AnimeInsSeg
+
+# 设置模型路径
+ckpt = r'models/AnimeInstanceSegmentation/rtmdetl_e60.ckpt'
+mask_thres = 0.3
+instance_thres = 0.3
+refine_kwargs = {'refine_method': 'refinenet_isnet'}  # 如果不使用 refinenet，设置为 None
+# refine_kwargs = None
+
+# 初始化模型
+net = AnimeInsSeg(ckpt, mask_thr=mask_thres, refine_kwargs=refine_kwargs)
+
+def has_instances(image_path):
+    # 读取图像
+    img = cv2.imread(image_path)
+    img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
+    
+    # 推理
+    instances = net.infer(
+        img,
+        output_type='numpy',
+        pred_score_thr=instance_thres
+    )
+
+    # 如果没有检测到对象，返回 False
+    if instances.bboxes is None:
+        return False
+    return True
+
+def copy_images_without_instances(input_path, output_dir):
+    input_path = Path(input_path)
+    output_dir = Path(output_dir)
+
+    if not output_dir.exists():
+        output_dir.mkdir(parents=True)
+
+    image_paths = list(input_path.rglob("*.png")) + list(input_path.rglob("*.jpg"))
+    for image_path in tqdm(image_paths, desc="Processing images"):
+        if not has_instances(image_path):
+            # 拷贝不包含实例的图片到目标文件夹
+            shutil.copy(image_path, output_dir / image_path.name)
+
+def main():
+    parser = argparse.ArgumentParser(description="Copy images without instances to a target directory")
+    parser.add_argument("input_path", type=str, help="Path to the input image or folder")
+    parser.add_argument("output_dir", type=str, help="Path to the output directory")
+    args = parser.parse_args()
+
+    copy_images_without_instances(args.input_path, args.output_dir)
+
+if __name__ == "__main__":
+    main()

noinstance_v2i.py

@@ -0,0 +1,110 @@
+#### pip install imagehash
+#### python noinstance_v2i.py 翠星测试.mp4  翠星图片
+#### python noinstance_v2i.py 翠星测试.mp4  翠星图片25_0_1
+
+import os
+import cv2
+import argparse
+import shutil
+import imagehash
+from pathlib import Path
+from tqdm import tqdm
+from moviepy.editor import VideoFileClip
+from collections import deque
+from PIL import Image
+
+from animeinsseg import AnimeInsSeg
+
+# 配置参数
+CKPT_PATH = "models/AnimeInstanceSegmentation/rtmdetl_e60.ckpt"
+DOWNSAMPLE_SIZE = (320, 180)  # 下采样尺寸提升处理速度
+MIN_SCENE_CHANGE = 0.3        # 场景变化阈值 (0-1)
+HASH_THRESHOLD = 25            # 哈希相似度阈值
+FRAME_BUFFER = 10             # 连续帧缓冲区
+
+class VideoProcessor:
+    def __init__(self):
+        self.net = AnimeInsSeg(
+            CKPT_PATH, 
+            mask_thr=0.1,
+            refine_kwargs={'refine_method': 'refinenet_isnet'}
+        )
+        self.hash_dict = {}
+        self.last_saved_hash = None
+
+    def _preprocess_frame(self, frame):
+        """预处理帧：下采样 + 灰度化"""
+        small_frame = cv2.resize(frame, DOWNSAMPLE_SIZE)
+        return cv2.cvtColor(small_frame, cv2.COLOR_BGR2GRAY)
+
+    def _has_human(self, frame):
+        """快速人物检测"""
+        instances = self.net.infer(
+            cv2.cvtColor(frame, cv2.COLOR_BGR2RGB),
+            output_type='numpy',
+            pred_score_thr=0.5  # 适当提高置信度阈值加速判断
+        )
+        return instances.bboxes is not None
+
+    def _is_duplicate(self, frame):
+        """基于感知哈希的去重检测"""
+        current_hash = imagehash.dhash(Image.fromarray(frame))
+        for existing_hash in self.hash_dict.values():
+            if current_hash - existing_hash < HASH_THRESHOLD:
+                return True
+        return False
+
+    def process_video(self, video_path, output_dir):
+        """核心处理流程"""
+        clip = VideoFileClip(str(video_path))
+        output_path = Path(output_dir)
+        output_path.mkdir(parents=True, exist_ok=True)
+
+        # 初始化场景检测
+        prev_frame = None
+        frame_buffer = deque(maxlen=FRAME_BUFFER)
+
+        for i, frame in enumerate(clip.iter_frames()):
+            # 转换为OpenCV格式
+            cv_frame = cv2.cvtColor(frame, cv2.COLOR_RGB2BGR)
+            
+            # 动态抽帧策略
+            processed_frame = self._preprocess_frame(cv_frame)
+            if prev_frame is not None:
+                # 使用结构相似性检测场景变化
+                similarity = cv2.compareHist(
+                    cv2.calcHist([processed_frame], [0], None, [256], [0, 256]),
+                    cv2.calcHist([prev_frame], [0], None, [256], [0, 256]),
+                    cv2.HISTCMP_CORREL
+                )
+                if similarity > (1 - MIN_SCENE_CHANGE):
+                    continue
+            
+            prev_frame = processed_frame
+            
+            # 缓冲区去重检测
+            if any(cv2.absdiff(processed_frame, f).sum() < 1000 for f in frame_buffer):
+                continue
+            frame_buffer.append(processed_frame)
+
+            # 执行人物检测
+            if not self._has_human(cv_frame):
+                # 哈希去重检查
+                if not self._is_duplicate(frame):
+                    frame_hash = imagehash.dhash(Image.fromarray(frame))
+                    self.hash_dict[i] = frame_hash
+                    cv2.imwrite(str(output_path / f"frame_{i:06d}.jpg"), cv_frame)
+
+        clip.close()
+
+def main():
+    parser = argparse.ArgumentParser(description="提取视频中无人物出现的帧")
+    parser.add_argument("video_path", type=str, help="输入视频路径")
+    parser.add_argument("output_dir", type=str, help="输出目录路径")
+    args = parser.parse_args()
+
+    processor = VideoProcessor()
+    processor.process_video(args.video_path, args.output_dir)
+
+if __name__ == "__main__":
+    main()