How to use it to improve accuracy of my already trained MVD video classifier?

[gigasurgeon]

I have trained an MVD video classifier. How can I use this to improve the f1 score or accuracy of my classifier?

[chien1706]

I have the same question when I want to apply this to my segmentation model. How can I apply this.

[tcourat]

I guess you need to train a new segmentation head on top of the upsampled features from a frozen backbone ?

[coolteemf]

I guess you need to train a new segmentation head on top of the upsampled features from a frozen backbone ?

I think the same, in my experiments with FeatUp I found it hard to adapt the decoder or head part of the network to the new features without retraining. I think most networks architecture are not really resolution independent, and downscaling the output of FeatUp would make no sense.

[gigasurgeon]

It is a classification model. Vit based. Will featup help in classification tasks?