Skip to content

Latest commit

 

History

History
36 lines (26 loc) · 1.09 KB

bundleGPB.md

File metadata and controls

36 lines (26 loc) · 1.09 KB

Bundle Adjustment on a Graph Processor

Author: Joseph Ortiz & Andrew Davison

Year: 2020

Notes:

  • IPU = new calcul unit for AI with massive parallelisation
  • GPB = general case of loopy belief prop.
  • Not competitive on CPU but take
  • BP = algorithm for computing marginal from joint probability
  • When distributions are Gaussians, LGBP converges to the correct marginal

BA factor graph:

  • 2 factors: measurement factor and prior factor
  • prior factor = comparison to a prior distribution
  • prior errors are required to set scale
  • linearization of the projection function
  • parametrization with information matrix and vector for measurement model

GPB:

  • variable nodes update their belief by taking a product of incoming messages from adjacent factors
  • synchronous scheduling: all factor nodes relinearise and send messages at each iterations
  • message damping to stabilise the convergence

Results:

  • Metrics to compare BA is ARE (average reprojection error)

Commentaires:

Pas appris grd chose de nouveau, revoir la notion de marginalisation. Idée appliquer GPB à la fusion de capteurs