About VGGSound Preparation

[misoyuri]

Hi, thanks for your contribution to audio source localization.

Now, I am trying to train your model FANC with VGGSound.
I have the VGGSound dataset but I do not know how to prepare the dataset like your 'prepare_flickr_soundnet.py'

It would be very appreciated if you share the preparation codes for VGGSound.

Also, are there any experimental results about utilizing Resnet50 or a bigger model, not Resnet18?

Thank you.

[weixuansun]

Hi, please follow https://github.com/hche11/Localizing-Visual-Sounds-the-Hard-Way for data preparation.
For model size, we observe that ResNet18 has better results than ResNet50, and large models are likely to be overfitting.

[misoyuri]

Thank you for sharing!

If you have time to share codes, please share the preprocessing codes for the VGGSound dataset.
I checked the repo you mentioned but could not find any codes about that.