How to train on ActivityNet dataset about rskp HOT 1 CLOSED

    "dropout":0.7,
    "lr":2e-5,
    "weight_decay":0.001,
    "inp_feat_num":2048,
    "out_feat_num":2048,
    "scale_factor":10.0,     # scaling factor for calculating classification scores
    "n_mu":8,     # the number of GMM centers
    'em_iter':2,      # the number of EM iterations
    "o_weight":0.8,      # the weight of main branch for TCAM fusion during testing
    "m_weight":0.2,     # the weight of intra-video branch for TCAM fusion during testing
    "lambda_b":0.1,     # weight of the multiple instance learning head of the classification head (the weight of the Class-agnostic attention head is set as 1.0)
    "lambda_att":0.1,     # weight of attention normalization loss
    "lambda_spl":1.0,     # weight of pseudo label loss