You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Thank you for the great work, though I have a question about the modality_offset, did you use that during the noise injection training? If no, the only modulation to the embeddings is the noise from N(0, variance), where variance is calculated from multiple samples, is that right?
The text was updated successfully, but these errors were encountered:
Hi, thanks for the interest in our work. We describe in the paper both with and without modalities offset. When it is without the offset so yes, it's just the variance that needs to be estimated using a few samples.
Thank you for the great work, though I have a question about the modality_offset, did you use that during the noise injection training? If no, the only modulation to the embeddings is the noise from N(0, variance), where variance is calculated from multiple samples, is that right?
The text was updated successfully, but these errors were encountered: