yc2_tsn_pdvcl_gt.yml

id: yc2_tsn_pdvcl_gt

visual_feature_type: ['resnet', 'bn']
visual_feature_folder: ['data/yc2/features/resnet_bn/', 'data/yc2/features/resnet_bn/']
feature_dim: 3072
invalid_video_json: []
train_caption_file: 'data/yc2/captiondata/yc2_train.json'
val_caption_file: 'data/yc2/captiondata/yc2_val.json'
gt_file_for_eval: ['data/yc2/captiondata/yc2_val.json']
gt_file_for_para_eval: ['data/yc2/captiondata/para/para_yc2_val.json']
dict_file: data/yc2/vocabulary_youcook2.json
vocab_size: 1607

train_proposal_type: gt
gt_proposal_sample_num: 30
sample_method: nearest

batch_size: 1
lr: 0.00005
learning_rate_decay_start: 8
learning_rate_decay_every: 3
learning_rate_decay_rate: 0.5
weight_decay: 0.0001
save_all_checkpoint: 0

num_queries: 100
dec_layers: 2
enc_layers: 2
transformer_ff_dim: 512
transformer_dropout_prob: 0.1
frame_embedding_num: 200
caption_decoder_type: light
att_hid_size: 0

#with_box_refine: 1

fix_xcw: 1
set_cost_caption: 0
set_cost_giou: 4
set_cost_bbox: 0.0001
set_cost_class: 0
#cost_alpha: 0.5
#cost_gamma: 1
#focal_alpha: 0.5
#focal_gamma: 1
caption_loss_coef: 2
giou_loss_coef: 0
bbox_loss_coef: 0
cls_loss_coef: 0
count_loss_coef: 0
#max_eseq_length: 10
#lloss_cross_entropy: 0
#lloss_focal_loss: 0
#lloss_gau_mask: 1

#two_stage: 1
transformer_input_type: gt_proposals