Question about paper [# model parameter]

[chwoong]

Dear author,
Thank you so much for your work and code.
I have a question about the number of model parameters.

As I understand the paper, pixels are classified as the closest prototype among CK prototypes at inference time. In the end, we have to store CK prototypes, then I wonder why we don't interpret them as model parameters. Also, the number of prototypes to be stored is proportional to the number of classes. Is it just convention?

Thank you.

[wenguanwang]

@chwoong

The prototypes are viewed as non-learnable parameters, as they are computed as the mean of a group of feature representations. The parameters of the feature extractors are learnable parameters, which are learned by gradient optimization.

[chwoong]

@wenguanwang Thank you.

[chwoong]

@wenguanwang https://youtu.be/Jm2wKObfES0?t=285. Then, the figure below in your presentation video only considered learnable parameters, right?

[wenguanwang]

@chwoong Yes, here we mean learnable parameters : -)

[chwoong]

@wenguanwang thank you for quick response. have a good day:)

[Fly-Yun-Maker]

@wenguanwang @tfzhou Thank you for your excellent contributions, can you please tell me if the paper code only provides profiles for prototype learning on the Cityscapes dataset?