wgcban / ddpm-cd

Remote Sensing Change Detection using Denoising Diffusion Probabilistic Models
https://www.wgcban.com/research#h.ar24vwqlm021
MIT License
247 stars 29 forks source link

about the meaning of the feature images' names #20

Closed Garsonjw closed 1 year ago

Garsonjw commented 1 year ago

Hi, I am interested in your work. And I have a question about the meaning of the feature images' names. I got many feature images. And I want to know the meaning of the 'i' and the 'level'. Is it possible to map it to the 'multi-scale feature'? image

wgcban commented 1 year ago

@Garsonjw Hi, the level indicates the different levels of the unet and i indicate the different features of the corresponding level. For an example, Unet computes features at multiple scales like HxW, H/2 x W/2, so on. These are the different levels. At a given level, we get a set of features.

Garsonjw commented 1 year ago

thank you for your reply. The level of the feats is eleven(0 1 2 3 4 5 6 7 8 9 10). It is the number of ResnetBlocWithAttn in the upsampling module. But I do't konw which level HxW, H/2 x W/2, so on represents. In other words, how can I get the same images as Fig3.

wgcban commented 1 year ago

@Garsonjw yes, that is true. What you can do is print the shape of the feats for a given level. That way you know the spatial resolutions of feats. Alternatively you may utilize this: https://github.com/wgcban/ddpm-cd/blob/b0213c0049bab215e470326d97499ae69416a843/model/cd_modules/cd_head_v2.py#L15-L28