Vision KAN

We try to explore the application of KAN in visual tasks

we can use "KanMLPMixer","KanPermutator" to get a classification model:

python train_cifar10.py

We circumvent the problem that the classification model can only support fixed-size inputs by applying kan to the channel dimension or using windows in the spatial dimension.

we can use "(channel)kanSSR","SwinPermutatorKan","SwinConvKan" to get a Hyperspectral Image Restoration model / spectral super resolution model / semantic segmentation model:

cd ./fastkan/HyperSpectralmodel
python kan_fit_test.py.py

super spectral image task's training code is coming soon!

To do

semantic segmentation

thansk:

KAN, fast kan, MLP Mixer