config.py

import argparse
import datetime
import json
import random
import time
import numpy as np


def str2bool(string):
    return True if string.lower() == 'true' else False


def get_args_parser():
    parser = argparse.ArgumentParser('Set IDU Online Detector', add_help=False)
    parser.add_argument('--lr', default=1e-4, type=float)     # 1e-4
    parser.add_argument('--batch_size', default=128, type=int)
    parser.add_argument('--weight_decay', default=1e-4, type=float)
    parser.add_argument('--epochs', default=5, type=int)
    parser.add_argument('--resize_feature', default=False, type=str2bool, help='run resize prepare_data or not')
    parser.add_argument('--lr_drop', default=1, type=int)
    parser.add_argument('--clip_max_norm', default=1., type=float,
                        help='gradient clipping max norm')  # dataparallel
    parser.add_argument('--dataparallel', action='store_true', help='multi-gpus for training')
    parser.add_argument('--removelog', action='store_true', help='remove old log')

    # * Network
    parser.add_argument('--version', default='v3', type=str,
                        help="fixed or learned")  # learned  fixed
    # decoder
    parser.add_argument('--query_num', default=8, type=int,
                        help="Number of query_num (prediction)")
    parser.add_argument('--decoder_layers', default=5, type=int,
                        help="Number of decoder_layers")
    parser.add_argument('--decoder_embedding_dim', default=1024, type=int,   # 1024
                        help="decoder_embedding_dim")
    parser.add_argument('--decoder_embedding_dim_out', default=1024, type=int,  # 256 512 1024
                        help="decoder_embedding_dim_out")
    parser.add_argument('--decoder_attn_dropout_rate', default=0.1, type=float,  # 0.1=0.2
                        help="rate of decoder_attn_dropout_rate")
    parser.add_argument('--decoder_num_heads', default=4, type=int,  # 8 4
                        help="decoder_num_heads")
    parser.add_argument('--classification_pred_loss_coef', default=0.5, type=float)  # 0.5

    # encoder
    parser.add_argument('--enc_layers', default=64, type=int,
                        help="Number of enc_layers")
    parser.add_argument('--lr_backbone', default=1e-4, type=float,    # 2e-4
                        help="lr_backbone")
    parser.add_argument('--feature', default='Anet2016_feature_v2', type=str,
                        help="feature type")
    parser.add_argument('--dim_feature', default=3072, type=int,
                        help="input feature dims")
    parser.add_argument('--patch_dim', default=1, type=int,
                        help="input feature dims")
    parser.add_argument('--embedding_dim', default=1024, type=int,  # 1024
                        help="input feature dims")
    parser.add_argument('--num_heads', default=8, type=int,
                        help="input feature dims")
    parser.add_argument('--num_layers', default=3, type=int,
                        help="input feature dims")
    parser.add_argument('--attn_dropout_rate', default=0.1, type=float,
                        help="attn dropout")
    parser.add_argument('--positional_encoding_type', default='learned', type=str,
                        help="fixed or learned")  # learned  fixed

    parser.add_argument('--hidden_dim', default=1024, type=int,  # 512 1024
                        help="Size of the embeddings")
    parser.add_argument('--dropout_rate', default=0.1, type=float,
                        help="Dropout applied ")

    parser.add_argument('--numclass', default=22, type=int,
                        help="Number of class")

    # * Loss coefficients
    parser.add_argument('--classification_x_loss_coef', default=0.3, type=float)
    parser.add_argument('--classification_h_loss_coef', default=1, type=float)
    parser.add_argument('--similar_loss_coef', default=0.1, type=float)   # 0.3
    parser.add_argument('--margin', default=1., type=float)

    # dataset parameters
    parser.add_argument('--dataset_file', type=str, default='data/data_info_new.json')
    parser.add_argument('--frozen_weights', type=str, default=None)
    parser.add_argument('--thumos_data_path', type=str, default='/home/dancer/mycode/Temporal.Online.Detection/'
                                                                'Online.TRN.Pytorch/preprocess/')
    parser.add_argument('--thumos_anno_path', type=str, default='data/thumos_{}_anno.pickle')
    parser.add_argument('--remove_difficult', action='store_true')
    parser.add_argument('--device', default='cuda',
                        help='device to use for training / testing')

    parser.add_argument('--output_dir', default='models',
                        help='path where to save, empty for no saving')
    parser.add_argument('--seed', default=20, type=int)
    parser.add_argument('--resume', default='', help='resume from checkpoint')
    parser.add_argument('--start_epoch', default=1, type=int, metavar='N',
                        help='start epoch')

    parser.add_argument('--eval', action='store_true')
    parser.add_argument('--num_workers', default=8, type=int)

    # distributed training parameters
    parser.add_argument('--world_size', default=1, type=int,
                        help='number of distributed processes')
    parser.add_argument('--dist_url', default='tcp://127.0.0.1:12342', help='url used to set up distributed training')
    # 'env://'
    return parser