iFRAME (inhomogeneous Filters Random Field And Maximum Entropy)

Experiment 3: Unsupervised alignment of objects with unknown rotations, locations and scales

code and data for unsupervised alignment of Cats, Deers, Ducks, Pandas, and Sheep

Parameter Setting

(1) Cat

training images with window locator

cropped and aligned objects



(2) Deer

training images with window locator

cropped and aligned objects



(3) Duck

training images with window locator

cropped and aligned objects



(4) Panda

training images with window locator

cropped and aligned objects



(5) Sheep

training images with window locator

cropped and aligned objects



General Parameters: nOrient = 16; sizeTemplatex=100; sizeTemplatey=100; numIterationEM=2; LocalNormOrNot = true;
Detection Parameters: rotateShiftLimit = 3; numResolution = 7; scaleStepSize = 0.1; flipOrNot=1 only for Duck category ;
HMC Parameters: lambdaLearningRate = 0.01; epsilon = 0.01; L = 10; nIteraton = 5; nTileRow = 4; nTileCol = 4; numSample =3;