About the method of DenseTNT prediction

[HB1109]

Hello, I have trained and tested the map, and completed the merging, how to train DenseTNT

[alfredgu001324]

Thank you for your interest in our work! here is a script that you can use to do the training.

And if I remember correctly, you can modify this line to select whether you use the uncertainty input or not.

Hope this helps!

[HB1109]

I encountered the "predicted_map" missing, I read your previous solution, but I don't quite understand, scene-15190.pkl is complete in my code. Could you be a little more specific The command I'm running is‘’ epochs=10 batch=16 lr=0.0005 dropout=0.1 output_dir="" # output_dir where model is stored train_dir=/home/data/HiVT_stream/train/data/ # train data dir val_dir=/home/data/HiVT_stream/val/data/ # val data dir python src/run.py --nuscenes --future_frame_num 30 --do_train --data_dir $train_dir --data_dir_for_val $val_dir --output_dir $output_dir --hidden_size 128 --train_batch_size $batch --use_map --core_num 16 --use_centerline --distributed_training 1 --other_params semantic_lane direction l1_loss goals_2D enhance_global_graph subdivide goal_scoring laneGCN point_sub_graph lane_scoring complete_traj complete_traj-3 --eval_params optimization MRminFDE=0.0 cnt_sample=9 opti_time=0.1 --learning_rate $lr --hidden_dropout_prob $dropout --num_train_epochs $epochs for i in {1..$epochs} do echo $(python src/run.py --nuscenes --future_frame_num 30 --do_eval --data_dir $train_dir --data_dir_for_val $val_dir --output_dir $output_dir --hidden_size 128 --train_batch_size $batch --use_map --core_num 16 --use_centerline --distributed_training 1 --other_params semantic_lane direction l1_loss goals_2D enhance_global_graph subdivide goal_scoring laneGCN point_sub_graph lane_scoring complete_traj complete_traj-3 --eval_params optimization MRminFDE=0.0 cnt_sample=9 opti_time=0.1 --learning_rate $lr --hidden_dropout_prob $dropout --model_recover_path $i) >> $output_dir/eval_results done ‘’I only changed the path