what dose the 'SeqL' mean? about opengait

[chuanfushen]

I recommend you print out the seqL, you will feel about it.

[aleeyang]

thx, But there is nothing output in the terminal when I use print such asprint(SeqL)
There is only training info in the terminal .
I use vscode to run the code in remote server.
Is there a predefined debug method(print the variable) in OpenGait?

[ChaoFan996]

Yet, thanks for your suggestions.
At the training phase, the value of seqL should be None unless the input sequence's length is unfixed.
So, you can set sample_type as unfixed_ordered and have a try again.
Moreover, at the test phase, the sample_type is set to all_ordered in default,
in another word, you should print out seqL at this time to feel it.
(all means the unfixed length of real-world gait sequence).

[aleeyang]

thank you

[aleeyang]

Thank you again your your reply.I compare the original"Gaitset" source code with yours.
The set_pooling is quite different.
The code in the original is https://github.com/AbnerHqC/GaitSet/blob/5535943428b66415530d8379b648b8f74a294219/model/network/gaitset.py#L47
Even the batch_frame euqual to None, it use if self.batch_frame is None: return torch.max(x, 1) the max is opeated in the 1 dimension of the x

But your code is just just max,return self.pooling_func(seqs, **kwargs) the default dimension is 0

OpenGait/lib/modeling/modules.py

Line 51 in 6982c35

class PackSequenceWrapper(nn.Module):

[aleeyang]

I mean if the 'SeqL' in your code and 'bacth_frame' in the original code both point the same subject.When it is None,your code use max in the 0 dim which is the batch_size dimension,it seems wrong....
The original one use max in the 1 dimension(the sequence dimension) whether it is None or not
The set pooling is same as the frame pooling which is used to aggrate the temporal info right?
Maybe my understanding is wrong,hope you can point it out.
Thank you again!

[ChaoFan996]

Yet, **kwargs can pass the dim=1 into self.pooling_func (you call max here).