Class: Gsplat::Autograd::Function

Inherits:

Object

• Object
• Gsplat::Autograd::Function

Defined in:

lib/gsplat/autograd/function.rb

Overview

Base class for differentiable coarse-grained operations.

Direct Known Subclasses

Ops::Accumulate, Ops::AddClampMin, Ops::CameraBroadcast, Ops::CameraDirections, Ops::ConcatCoefficients, Ops::ConcatDepth, Ops::ConcatFeatures, Ops::DepthFeatures, Ops::Eval3dRasterize, Ops::Exp, Ops::FeatureSlice, Ops::FullyFusedProjection, Ops::Multiply, Ops::NormalizeDepth, Ops::NormalizeQuaternion, Ops::QuatScaleToCovarPreci, Ops::RasterizeToPixels, Ops::Sigmoid, Ops::SphericalHarmonics, Training::ImageFitter::MeanSquaredError, Training::Losses::L1, Training::Losses::Reconstruction, Training::Losses::RegularizedReconstruction, Training::Losses::StructuralSimilarity

Class Method Summary

• .apply(*inputs) ⇒ Variable⁺

  Executes forward and records a graph node when any input needs a gradient.

Class Method Details

.apply(*inputs) ⇒ Variable⁺

Executes forward and records a graph node when any input needs a gradient.

Parameters:

• inputs (Array<Object, Variable>)

Returns:

• (Variable, Array<Variable>)

# File 'lib/gsplat/autograd/function.rb', line 40

def apply(*inputs, **)
  needs_input_grad = inputs.map { |input| input.is_a?(Variable) && input.requires_grad? }
  context = Context.new(needs_input_grad, inputs)
  raw_inputs = inputs.map { |input| input.is_a?(Variable) ? input.data : input }
  result = forward(context, *raw_inputs, **)
  multiple_outputs = result.is_a?(Array)
  output_data = multiple_outputs ? result : [result]
  requires_grad = Autograd.grad_enabled? && needs_input_grad.any?
  node = GraphNode.new(self, context, inputs) if requires_grad
  outputs = output_data.map do |data|
    validate_output!(data)
    Variable.new(data, requires_grad: requires_grad, creator: node)
  end
  node.outputs = outputs if node

  multiple_outputs ? outputs : outputs.first
end