Kun Su

question about the srnn_loss

3

In the translate.py, when you test with srnn seeds, it looks like the srnn_loss you print out is only for the last action instead of an average loss of all...

Clarification re. Fixed Weight (FW) implementation

3

The intent of Fixed Weight (FW) and Fixed State (FS) strategies is to weaken the decoder such that the encoder final hidden state can be used as a robust representation...

MusicGen cross_attention=False?

1

I am confusing about the text2music config. It seems like in `config/model/lm/default.yaml` the `cross_attention=False` and I didn't see any overwrite in `musicgen_lm.yaml` but in the output of the fuser does...

how can we perform more data augmentation?

2

Hi, Sorry, I am a little bit new to this area. I am wondering how to augment the current HumanML3D data by simply concatenating the same or different motion sequences...

Hi, thanks for sharing the code! I have read the papers and watched the demo video. I find that the generated sound does temporally align with the visual information but...

Question on Masking ratio

5

Hi, thanks for sharing the Pytorch implementation! I am curious about how you select the stats for varied masking ratios. In the paper, you mentioned 'a truncated Gaussian distribution centered...

Hi, thanks for releasing the code. Since I am new to the acoustic literature, I may have a dumb question: how do you actually generate the IR (shown in Table...

Question for contrastive loss weight in the paper

3

I have a question regarding the weights used in CAV-MAE. It seems like the $\lambda_c$ could play an important role in the optimization. I understand it is due to the...

Hi, I am interested in the reason behind the large masking temperature used in Vampnet. I am working on a similar project and observe similar outcomes, e.g., larger masking temperature...