Can feature scale upsampling be singular before and after, such as x1=(1, 64, 80, 45) x2=(1, 64, 40, 23)

tiny-smart / dysample

(ICCV'23) Learning to Upsample by Learning to Sample

MIT License

112 stars 3 forks source link

Can feature scale upsampling be singular before and after, such as x1=(1, 64, 80, 45) x2=(1, 64, 40, 23) #15

Closed Prown666 closed 3 weeks ago

Prown666 commented 1 month ago

Can feature scale upsampling be singular before and after, such as x1=(1, 64, 80, 45) x2=(1, 64, 40, 23)

poppuppy commented 3 weeks ago

Conceptually it can, but, the current implementation can not. To obtain such a downsampler as you said, you can use x1 to generate sampling points of size (40, 23) and then resample x1 to x2.