minADE is not right

[shuishenbushui]

when I trained the Set Predictor, eval results are as follows:

other_errors {'stage_one_k': 3.0803698258030754, 'stage_one_recall': 0.9638398086076706, 'set_MR_pred': 0.0738903979863932, 'set_minFDE_pred': 1.3842987156892936}
{'minADE': 14.246385467933312, 'minFDE': 1.3842987156892916, 'MR': 0.0738903979863932}
ADE 14.316873007381048
DE@1 9.875993647573898
DE@2 19.429086244927365
DE@3 3.496420582410066

the minADE is absolutely wrong. I visualize the pred trajs, they are also not right, I think the complete traj module is not trained sufficient, but I trained the model follow your instructions, so where exactly is the problem?

[shuishenbushui]

Decoder.py 174, why you add a 'pass', skip the loss of pred trajs ？causing the complete traj module not trained.

[GentleSmile]

The 'pass' does not take any effect here. What is the value of 'complete_traj' in args.other_params, is it True?

[shuishenbushui]

Sorry，I made a mistake of 'pass'. But 'complete_traj' is indeed True. Can you give me a trained weight?

[shuishenbushui]

These are my parameters, Can you find what is wrong?

--argoverse
--future_frame_num
30
--do_train
--data_dir
train/data/
--output_dir
models_complete_traj
--model_recover_path=models.densetnt.set_predict.2/model_save/model.16.bin
--hidden_size
128
--train_batch_size
64
--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
set_predict=6
set_predict-6
data_ratio_per_epoch=0.4
set_predict-topk=0
set_predict-one_encoder
set_predict-MRratio=1.0
set_predict-train_recover=models.densetnt.set_predict.2/model_save/model.16.bin
--reuse_temp_file

[GentleSmile]

It seems that model_recover_path does not appear in the README

[shuishenbushui]

It doesn‘t matter，if I don't set model_recover_path, it will go to ./tmp to search the weight file. Can you give me a wechat number for deeper communication for details?

[GentleSmile]

What is your evaluation command? During evaluation, you should set model_recover_path to the correct ${OUTPUT_DIR}/model_save/model.16.bin.

[shuishenbushui]

Thank u, I knew my problem. model_recover_path should be set to weights of Set Predictor training, but set_predict-train_recover set should be set to weights of first training.