No explicit input argument to make train_dist_ada consume checkpoint of ap_recognition_network

[qwangku]

I am trying to repeat this step: Train policy network with well-trained ap recognition network. But it seems I cannot pass pre-trained checkpoint for ap network into train_dist_ada.py anywhere. Here is the exemplary command I am using:

python train_dist_ada.py --dataset activitynet --q_init 4 --backbone_net resnet_videoiq \
-d 50 -b 72 -j 16 --logdir <log_dir_path> --lr 0.0001 --epochs 50 --bit_width_family 4 2 \
--switch_bn --switch_clipval --frames_per_group 1 --groups 16 --loss_type KD_CE --q_weight-decay 0.0005 0.005 \
--multiprocessing-distributed --without_t_stride --disable_scaleup --p_lr 0.01 --efficient --balanced --efficient_w 0.16 \
--balanced_w 1 --resnet_imagenet_path <fp_recognition_network_path> \
--mobilenet_imagenet_path <mb_net_path> --p_epochs 50 --prefix e1b1_new --use_fp_as_bb \
--is_32fp --auto_resume --is_policy_pred --datadir <dataset_path> \

Could you provide some guidance on what is going on? Thanks.