zhangxiann / tpami_gait_identification Goto Github PK

您好，下载了您提供的预训练模型EV_Gait_3DGraph1.pkl后放到test_3d_graph.py测试时提示缺少一些键值，Missing key(s) in state_dict: "conv1.root.weight", "block1.left_conv1.root.weight", "block1.left_conv2.root.weight", "block1.shortcut_conv.root.weight", "block2.left_conv1.root.weight", "block2.left_conv2.root.weight", "block2.shortcut_conv.root.weight", "block3.left_conv1.root.weight", "block3.left_conv2.root.weight", "block3.shortcut_conv.root.weight".
请问是什么原因

hi, there. Thanks for your sharing.
When I found the link of "DVS-128-Night" is none.
Can you share the download link again?

您好
下载您提供的模型后，运行报错
AttributeError: module 'torch.nn' has no attribute 'UninitializedParameter'
请问是什么问题？

Hi, thanks for your great work.
In your paper 'EV-gait: Event-based robust gait recognition using dynamic vision sensors' you propose a noise cancellation module to remove noise via motion consistancy. But it seems that this module is removed in your TPAMI paper. Can you provide the code for this module?
Many thanks in advance.

From the link of the DVS-Gait-Night dataset you provided, I find the number of each object is under 200. So I want to know If there exists any difference between the dataset you upload from your description of this dataset.

下采样后的单个数据文件的事件信息数据量都为2048，但是matlab和python代码看起来无法控制事件信息数量为2048，想问一下如何控制事件信息数量

RT，请问大佬可不可以分享下

Hi, thanks for your contributions to the event-based vision. For the journal version of this work [paper], I have some questions about the Eq(3):

1). Does the " J Gaussian kernels" should be "K Gaussian kernels"?

2). Does the "p = 1, 2, 3, ... , P" in Eq(3) mean the index of each feature value of the P-dimensional feature vector?

3). "where $f_p^'$ is one entry of the P-dimensional output feature vector." But this term is not defined in Eq (3). It really confused me.

Thanks. Looking forward to your reply.

I am trying but it sends me to a link that i cannot understand anything

Hi,

When run EV-Gait-IMG/gait_cnn_tf.py, there is one file missing.

environments.py.

thanks,

首先对您的工作非常赞赏，这真的是一个很棒的工作！
利用数据集和您提供的方法测试了一下，test acc is 0.9525，看来应该很不错。
不过我想请问一下，有没有相应的可视化应用方案？
比如，输入两段视频，来验证是否为同一人的步态？
或者您给一些建议？再次感谢您的工作！

@zhangxiann