Questions about FDS

[ChengkaiYang]

Hi @YyzHarry, I noticed your excellent work and want to apply to my own work. My work is a multi-modal work. That is to say my encoded features fuse multi-modal information and want to use the feature to regress. Is FDS efficient in this case? Different from your example, the encoded features in my task contain more aspects but not only an image, does this influence FDS? I failed to use FDS in my task and noticed a phenomenon that the mean difference between bin features and smoothed bin features are getting smaller and smaller with the epoch increase, but the var difference between bin features and smoothed bin features are getting bigger and bigger with the epoch increase. Looking forward for your reply.Thank you so much!