Paper: Building Temporal Kernels with Orthogonal Polynomials
This is a self-contained repo for using temporal kernels parameterized by orthogonal polynomials. For example, it can be used as a drop-in replacement for convolutional layers (only supporting nn.Conv3d layers for now), where the last dimension (assumed to be temporal) will be parameterized by orthogonal polynomials up to a given degree.
from model import PleiadesLayer
layer = PleiadesLayer(2, 8, kernel_size=(3, 3, 20), degrees=4)The structured temporal kernels can also easily be resampled into different kernel sizes without needing to retrain the network.
layer.resample(10) # downsample the kernel size from 20 to 10