shuaishiliu / sgcn Goto Github PK

For example,loss =-0.002 and loss =-0.033,which better?

Hi,

Nice work,

I would like to get something to clarify about your results. When comparing to other methods on the ETH-UCY benchmark, do you consider multiple predicted trajectories (n=20), or does your method generate only a single trajectory?

I could not found that information on the paper.

Hi,

Thank you very much for your work and I'm very interested in trajectory prediction.

I have some problems about the visualization of predicted trajectory.
1. How to visualize the predicted trajectories distribution in your paper Fig4?
2. How to plot the trajectories in the scene picture just like the Fig5 in your paper?
Will you provide the code if you like? I hope you can help me solve the problem. Thank you very muh!

Best,
Jincan

····

Hello, I would like to try to reproduce this code and found that the code is incomplete, can you send me a copy of the complete code?

您好，

我有一下几点疑惑：
1、non_linear_ped以及loss_mask似乎并未使用到？
2、在处理数据时，应该只有连续在20帧都出现的行人的轨迹会被拿来训练，但是那些出现时长少于20帧的行人同样会对预测的轨迹产生影响，这个问题该怎么理解？
3、邻接矩阵是否是通过自注意力计算产生？

Hello, I would like to try to reproduce this code and found that the code is incomplete, can you send me a copy of the complete code?

Thanks for your great work!

I would like to try to reproduce your work, but the current version seems to be missing some necessary components. May I ask if you are planning to release your complete code again?

I really appreciate your understanding and look forward to your reply.

I'm interested in your work. Could you share your code for further development?

Hello, do you have any introduction to the preparation of the data set? I see your excellent work and want to refer to the training to have a look. Because the research direction is not in this field, I am not familiar with the data set. Looking forward to your early reply, thank you.

请问什么时候能公布代码？
When will the code be made public？

请问什么时候能公布代码？
When will the code be made public？

Hi,

Thanks for your great job and open source code! I am confused about the shortcut defined in class "AsymmetricConvolution". According to equation (2) in your paper, x2 should be the F(l). Then why return x2+shortcut?

Hello @shuaishiliu,

How can I convert the world coordinate to image coordinate please?

Can you share with us the code for visualization of figures 4 and 5 please?

Thanks

Hello @shuaishiliu,

When I tested the same code with your pretrained model (checkpoints), I hadn't the same results of the paper. Can you explain me why please?

Another question, how can I convert the matrix to cordinates and visualize the trajectories in figure 4 and 5 in the paper please? can you share with me the code for the visualization of the trajectories please?

Thanks

Thanks for your great work!

I would like to try to reproduce your work, but the current version seems to be missing some necessary components. May I ask if you are planning to release your complete code again?

I really appreciate your understanding and look forward to your reply.

training an epoch, the loss would be 0.02 or so, and then it would be negative. What kind of case is convergence?