Why you put the GRU in the loop of n_layers?

[wuhaoran111]

Dear author,
i have read your code like model_graph.py and i found that the GRUs for the pedestrian and context are in the loop of the n_layers of the GNN, the structure is like follows:

    for batch_size:
        for time_step:
            for gnn_layer:
                 curr_ped, h_ped = self.ped_gru(curr_ped, h_ped) # the GRU for the pedestrian
                 ......

In this structure the GRU updates time_step*gnn_layer steps. But in my perspective, to simulate the temporal relation, the GRU should only update time_step steps. And this makes me confused, could you give me an explanation?

[MrYu001]

Hello bro, have you run the code on the JAAD dataset, i think the JAAD dataset needs to be processed and saved as .npy file, but the author of this repository did not provide it.

[wuhaoran111]

sorry bro，we use another method instead, and i haven't tested in the JAAD dataset yet.

…

------------------ 原始邮件 ------------------ 发件人: "tomas ***@***.***>; 发送时间: 2022年1月21日(星期五) 上午10:36 收件人: ***@***.***>; 抄送: ***@***.***>; ***@***.***>; 主题: Re: [StanfordVL/STR-PIP] Why you put the GRU in the loop of n_layers? (#6) Hello bro, have you run the code on the JAAD dataset, i think the JAAD dataset needs to be processed and saved as .npy file, but the author of this repository did not provide it. — Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android. You are receiving this because you authored the thread.Message ID: ***@***.***>