def get_parser(): parser = argparse.ArgumentParser(description='Face detection and classification for politicians in Japanese TV.') # Important configuration variables parser.add_argument('--dataset', type=str, default='mot17', help='Mode name for saving files.') parser.add_argument('--mode', default='train', type=str, help='train or test.') parser.add_argument('--detector', type=str, default='YOLOX', help='Detector to be used. FRCNN, SDP, Bresee, SGT, YOLOX, GT.') parser.add_argument('--reid', type=str, default=None, help='Reidentification model to be used. SBS, MGN.') parser.add_argument('--mod', type=str, default=None, help='Tracker name modifier to do testing of features.') # Paths parser.add_argument('--datapath', type=str, default='datasets/MOT17Det', help='Dataset path with frames inside.') parser.add_argument('--feat', type=str, default='feats', help='Features files path.') # Tracking-specific configuration variables parser.add_argument('--max_iou_th', type=float, default=0.15, help='Max value to multiply the distance of two close objects.') parser.add_argument('--w_tracklet', type=int, default=10, help='Window size per tracklet') parser.add_argument('--w_fuse', type=int, default=3, help='Window size per fusion in hierarchy') parser.add_argument('--max_prop', type=int, default=10000, help='Difficult the fusion when the frame difference is larger than this value.') parser.add_argument('--fps_ratio', type=int, default=1, help='Use lower fps dataset if lower than 1.') # Flags parser.add_argument('--save_feats', action='store_true', help='Save tracking + feature vectors as pkl file for analysis.') parser.add_argument('--iou', action='store_true', help='Add IoU distance to further improve the tracker.') parser.add_argument('--temp', action='store_true', help='Use temporal distance to further improve the tracker.') parser.add_argument('--spatial', action='store_true', help='Use spatial distance to further improve the tracker.') parser.add_argument('--motion', action='store_true', help='Add motion estimation to further improve the tracker.') parser.add_argument('--randorder', action='store_true', help='Random order of lifted frames for testing.') parser.add_argument('--noncont', action='store_true', help='Do not enforce continuous clustering. Allow all tracklets to cluster with whoever they want.') return parser
时间: 2024-01-15 22:02:59 浏览: 172
这是一个 Python 脚本,通过 argparse 模块来接收命令行参数。这个脚本的作用是进行政治家在日本电视节目中的人脸检测和分类。其中包含了一些重要的配置变量,比如数据集名称、模式名称、检测器名称、重识别模型名称等等。此外,还有一些路径信息和跟踪器特有的配置变量,比如最大 IOU 阈值、窗口大小、帧速率等等。最后,还有一些标志用于控制程序的行为,比如是否保存特征向量、是否添加 IOU 距离等等。
阅读全文
相关推荐
















