如何提高视频分析的速度？

Question

我有一个超过1小时长的视频，并且有110000+帧。我正在使用下面的代码来分析视频中的面部表情。然而，整个过程太长，并且估计完成这项任务的时间是超过34小时！我的问题是:是否有任何方法可以加快这个过程(例如跳过帧)？每3帧分析一次？)

以下是代码：

from fer import Video
from fer import FER
import os
import sys
import pandas as pd

location_videofile = "/Users/M/Desktop/videoplayback.mp4"
input_video = Video(location_videofile)

processing_data = input_video.analyze(face_detector, display=False)

Answer 1

没有在fer的文档中显式地看到这一点，但是如果您查看Video对象的源代码，analyze会接受一堆关键字参数。

有关的双边投资条约是：

def analyze(
        self,
        detector, # fer.FER instance
        display: bool = False,
        output: str = "csv",
        frequency: Optional[int] = None,
        max_results: int = None,
        save_fps: Optional[int] = None,
        video_id: Optional[str] = None,
        save_frames: bool = True,
        save_video: bool = True,
        annotate_frames: bool = True,
        zip_images: bool = True,
        detection_box: Optional[dict] = None
    ) -> list:
        """Recognize facial expressions in video using `detector`.
        
        Args:
        
            detector (fer.FER): facial expression recognizer
            display (bool): show images with cv2.imshow
            output (str): csv or pandas
            frequency (int): inference on every nth frame (higher number is faster)
            max_results (int): number of frames to run inference before stopping
            save_fps (bool): inference frequency = video fps // save_fps
            video_id (str): filename for saving
            save_frames (bool): saves frames to directory
            save_videos (bool): saves output video
            annotate_frames (bool): add emotion labels
            zip_images (bool): compress output
            detection_box (dict): dict with bounding box for subimage (xmin, xmax, ymin, ymax)        

因此，您可以设置“使用frequency”来减少采样频率，例如，每3帧采样一次：

processing_data = input_video.analyze(face_detector, display=False, frequency=3)