Hello, may I ask if ATST Frame can publicly disclose the models and scripts used for inference?

[Angelalilyer]

I only found the training code, and my test dataset is unlabeled. I would like to try using ATST Frame to detect sound events in a new dataset. Perhaps you have inference code and models? Looking forward to your reply!

[lmaxwell]

Hi,
Currently, only self-supervised pretrained checkpoint are pulic in this repo. But we are happy to make the finetuned checkpoint public.
As you are targeting for detercting sound events, do you mean the checkpoint and inference code of ATST-Frame finetuned on strongly-labeled audioset?

[Angelalilyer]

Thanks for your reply~！
Yes, that is exactly what i need! So I wonder When you will release the “checkpoint and reference code of ATST Frame refined on strongly labeled audioset” ?

[SaoYear]

Hi, I would let you know once the uploading finished ; )

[SaoYear]

@Angelalilyer You could try this checkpoint file, hope it helps!

[Angelalilyer]

@Angelalilyer You could try this checkpoint file, hope it helps!

Thank you so much!!

[Angelalilyer]

@Angelalilyer You could try this checkpoint file, hope it helps!

Is there a complete inference code available? I tried to modify "audiossl/audiossl/methods/atstframe/downstream/train_strong. py" but kept reporting errors.
Sorry to bother you again～

[lmaxwell]

Did you solve the problem?

[Angelalilyer]

Did you solve the problem?

I tried to write inference code myself, but I couldn't output predicted labels，may I ask if there is a relatively complete inference code for ATST Frame (Audioset strong label)? Thanks for your help again!

[lmaxwell]

I write a quick solution in a new pull request #13 , can you test it ?

[Angelalilyer]

I write a quick solution in a new pull request #13 , can you test it ?

thanks!!
I can run this code, but there are some labels in the "Labels" list that I cannot find their corresponding names in the "ontology.json" of strongly-labeled audioset. For example, "/m/0c1tlg"

[SaoYear]

The strong AudioSet includes some extra Mids excluded by the original AudioSet ontology, you could refer to the official page and download the mid_to_display_name.tsv file.

According the the file, /m/0c1tlg refers to Electric rotor drone, quadcopter.

[Angelalilyer]

The strong AudioSet includes some extra Mids excluded by the original AudioSet ontology, you could refer to the official page and download the mid_to_display_name.tsv file.

According the the file, /m/0c1tlg refers to Electric rotor drone, quadcopter.

Thank you very much for your help! The problem has been solved～