search

JoeHEZHAO / expert_traj

ICCV2021 Expert-Goal Trajectory Prediction
25 stars 5 forks source link

Error happens in line 280 when running train_sdd.py #8

Open WangHonghui123 opened 1 year ago

WangHonghui123 commented 1 year ago

When I am running train_sdd.py, I find an error about the dimension of the data. The error is shown as follows:

WeChatbdea3d06d3a9e7498e5c3e5304612b77

And I find that there are some ndarray elements with DIFFERENT dimensions in traj_new list. For example, the dimension of traj_new[0] is (513, 20, 2), but the dimension of traj_new[4] is (512, 20, 2).

I hope you can solve this issue for me. Thank you very much.

JoeHEZHAO commented 1 year ago

Can you leave me with more information?

I am able to train it. Finished a training just now.

Screen Shot 2023-04-16 at 2 38 27 PM
WangHonghui123 commented 1 year ago

Thank you. I find that I am able to train SDD. It is very wired that it cannot be able to train at that time. Thanks. By the way, could you please provide the training model code for ETH/UCY dataset? Thank you.

JoeHEZHAO commented 1 year ago

@WangHonghui123 Sure, I can do that this weekend. Or you can

  1. Look in the dataloading code in test_ethucy.py
  2. Swap the dataloader in train_sdd.py

Everything else is the same I believe.

WangHonghui123 commented 1 year ago

@JoeHEZHAO Thank you for your suggestions.

I have some questions after I see your codes:

  1. The data processing in SDD dataset is different from that in ETH/UCY dataset
  2. The model used in train_sdd.py is different from that used in test_ethucy.py. In train_sdd.py, everything is same as what you mentioned in your paper. However, In test_ethucy.py, you use spatial temporal graph convolution from Social-STGCNN but you did not mention and elaborate it in your paper.

I am looking forward to your answers and your training script for ETH/UCY dataset. Have a nice day.