norse.torch.module.receptive_field

These receptive fields are derived from scale-space theory, specifically in the paper Normative theory of visual receptive fields by Lindeberg, 2021.

For use in spiking / binary signals, see the paper on Translation and Scale Invariance for Event-Based Object tracking by Pedersen et al., 2023

Classes

ParameterizedSpatialReceptiveField2d(...[, ...])	A parameterized version of the SpatialReceptiveField2d module, where the scales, angles, ratios, x and y are optimized and updated for each kernel individually during training.
SampledSpatialReceptiveField2d(in_channels, ...)	Creates a spatial receptive field as 2-dimensional convolutions, sampled over a set of scales, angles, ratios, x, y and derivatives.
SpatialReceptiveField2d(in_channels, size, ...)	Creates a spatial receptive field as 2-dimensional convolutions.
TemporalReceptiveField(shape, n_scales, ...)	Creates n_scales temporal receptive fields for arbitrary n-dimensional inputs.