Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

shape mismatch problem in input_feature.py #17

Closed
yangalan123 opened this issue May 15, 2018 · 8 comments
Closed

shape mismatch problem in input_feature.py #17

yangalan123 opened this issue May 15, 2018 · 8 comments

Comments

@yangalan123
Copy link

yangalan123 commented May 15, 2018

Hello! We are trying to make our own input pipeline. However, when we follow the getitem method in Audioset (with the setting that cube_shape is (20,80,40)), there is a shape mismatch when the model tries to feed data for batch_speech (placeholder with the shape of (20,80,.40,1)).

After carefully review the code in train_softmax.py, we find that the input shape will conflict with the transpose operation in following code:

speech_train = np.transpose(speech_train[None, :, :, :, :], axes=(1, 4, 2, 3, 0))

What is the solution? Could you give us any help?

@astorfi
Copy link
Owner

astorfi commented May 16, 2018

Thank you for your question, the getitem is just for showcasing the deadline and all the shapes that you are mentioning are just my custom way of inputting data. The pipeline must by modify and adapted considering your desired architecture and etc. You can create the pipeline in a way that you don't even need np.transpose which makes the code slow for sure. Please go to the details of the provided class for input pipeline to grasp an idea about the details.

@yangalan123
Copy link
Author

Sorry, but I didn't see a example class for input pipeline in the current commit(Do you mean input_feature.py?) However, I do find make_development.py in the pase commit line. Could we just use it? Are there some known bugs in make_development.py which makes you delete the file?

Thanks for your patience. Looking forward for your reply.

@astorfi
Copy link
Owner

astorfi commented May 16, 2018

If you are using an old fork please replace it with a new one. Yes, the input_feature.py function is pretty much everything you need for the input pipeline.

@chenhaishun
Copy link

@yangalan123 hi~
I am also studying this project. I hope to have some exchanges with you.please~[QQ:540851498]

@yangalan123
Copy link
Author

yangalan123 commented May 20, 2018

@chenhaishun Thank you for your kindness, but honestly I don't make detailed researches about this field. I only want to make some application on the basis of this great paper. So...I am afraid I cannot help you much.

@yangalan123
Copy link
Author

@astorfi Thank you for your kind help, I have found out a possible solution to incorporate input_feature.py into my project. I will close this issue in advance. If anyone has the similar questions, please reopen.

@chenhaishun
Copy link

Hi~
We can also communicate in other areas. To be honest in this area, I have only just started, and I still hope to share my experience with you.
So...please

@0ifeng0
Copy link

0ifeng0 commented May 28, 2018

@yangalan123 I have the same problem with the shape mismatch.Can you show your sulution?Thank you.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

4 participants