iFRAME (inhomogeneous Filters Random Field And Maximum Entropy)

Experiment 2.2: Sparse FRAME (Orthogonal Shared Matching Pursuit)

Code and dataset

Extra results

Contents
Case 1 : Cat
Case 2 : Wolf
Case 3 : Pigeon
Case 4 : Zebra
Case 5 : Deer
Case 6 : Flamingo
Case 7 : Hummingbird
Case 8 : Cougar
Case 9 : Lion

Case 1: Cat

Step 1: Sparsification: reconstruction-based filters selection by orthogonal matching pursuit (arg-max for deformation modeling)

learned templates (Layout of Gabors and DoGs)

deformed templates at different scales

template

 (i) original (ii) templates (iii) reconstruced (iv) residual

Step 2: Simulation: synthesis by learning Frame model (arg-max for deformation modeling)

parameters setting:

General Parameters: nOrient = 16; sizeTemplatex=100; sizeTemplatey=100; GaborScaleList=[ 1.4, 1, 0.7, 0.5]; DoGScaleList =[18.90, 13.36]; LocationShiftLimit=3; OrientShiftLimit=1; #sketches=300;
HMC Parameters: lambdaLearningRate = 0.1/sqrt(sigsq); epsilon = 0.03; L = 10; nIteraton =100; 6x6 chains; sigsq=10;

Case 2: Wolf

Step 1: Sparsification: reconstruction-based filters selection by matching pursuit (arg-max for deformation modeling)

learned templates (Layout of Gabors and DoGs)

deformed templates at different scales

template

 (i) original (ii) templates (iii) reconstruced (iv) residual

Step 2: Simulation: synthesis by learning Frame model (arg-max for deformation modeling)

parameters setting:

General Parameters: nOrient = 16; sizeTemplatex=100; sizeTemplatey=100; GaborScaleList=[ 1.4, 1, 0.7, 0.5]; DoGScaleList =[18.90, 13.36]; LocationShiftLimit=3; OrientShiftLimit=1; #sketches=300;
HMC Parameters: lambdaLearningRate = 0.1/sqrt(sigsq); epsilon = 0.03; L = 10; nIteraton =100; 6x6 chains; sigsq=10;

Case 3: Pigeon

Step 1: Sparsification: reconstruction-based filters selection by matching pursuit (arg-max for deformation modeling)

learned templates (Layout of Gabors and DoGs)

deformed templates at different scales

template

 (i) original (ii) templates (iii) reconstruced (iv) residual

Step 2: Simulation: synthesis by learning Frame model (arg-max for deformation modeling)

parameters setting:

General Parameters: nOrient = 16; sizeTemplatex=95; sizeTemplatey=100; GaborScaleList=[ 1.4, 1, 0.7, 0.5]; DoGScaleList =[18.90, 13.36]; LocationShiftLimit=3; OrientShiftLimit=1; #sketches=250;
HMC Parameters: lambdaLearningRate = 0.1/sqrt(sigsq); epsilon = 0.03; L = 10; nIteraton =100; 6x6 chains; sigsq=10;

Case 4: Zebra

Step 1: Sparsification: reconstruction-based filters selection by matching pursuit (arg-max for deformation modeling)

learned templates (Layout of Gabors and DoGs)

deformed templates at different scales

template

 (i) original (ii) templates (iii) reconstruced (iv) residual

Step 2: Simulation: synthesis by learning Frame model (arg-max for deformation modeling)

parameters setting:

General Parameters: nOrient = 16; sizeTemplatex=103; sizeTemplatey=123; GaborScaleList=[ 1.4, 1, 0.7, 0.5]; DoGScaleList =[18.90, 13.36]; LocationShiftLimit=3; OrientShiftLimit=1; #sketches=300;
HMC Parameters: lambdaLearningRate = 0.1/sqrt(sigsq); epsilon = 0.03; L = 10; nIteraton =100; 6x6 chains; sigsq=10;

Case 5: Deer

Step 1: Sparsification: reconstruction-based filters selection by matching pursuit (arg-max for deformation modeling)

learned templates (Layout of Gabors and DoGs)

deformed templates at different scales

template

 (i) original (ii) templates (iii) reconstruced (iv) residual

Step 2: Simulation: synthesis by learning Frame model (arg-max for deformation modeling)

parameters setting:

General Parameters: nOrient = 16; sizeTemplatex=100; sizeTemplatey=110; GaborScaleList=[ 1.4, 1, 0.7, 0.5]; DoGScaleList =[18.90, 13.36]; LocationShiftLimit=3; OrientShiftLimit=1; #sketches=300;
HMC Parameters: lambdaLearningRate = 0.1/sqrt(sigsq); epsilon = 0.03; L = 10; nIteraton =100; 6x6 chains; sigsq=10;

Case 6: Flamingo

Step 1: Sparsification: reconstruction-based filters selection by matching pursuit (arg-max for deformation modeling)

learned templates (Layout of Gabors and DoGs)

deformed templates at different scales

template

 (i) original (ii) templates (iii) reconstruced (iv) residual

Step 2: Simulation: synthesis by learning Frame model (arg-max for deformation modeling)

parameters setting:

General Parameters: nOrient = 16; sizeTemplatex=127; sizeTemplatey=100; GaborScaleList=[ 1.4, 1, 0.7, 0.5]; DoGScaleList =[18.90, 13.36]; LocationShiftLimit=3; OrientShiftLimit=1; #sketches=300;
HMC Parameters: lambdaLearningRate = 0.1/sqrt(sigsq); epsilon = 0.03; L = 10; nIteraton =100; 6x6 chains; sigsq=10;

Case 7: Hummingbird

Step 1: Sparsification: reconstruction-based filters selection by matching pursuit (arg-max for deformation modeling)

learned templates (Layout of Gabors and DoGs)

deformed templates at different scales

template

 (i) original (ii) templates (iii) reconstruced (iv) residual

Step 2: Simulation: synthesis by learning Frame model (arg-max for deformation modeling)

parameters setting:

General Parameters: nOrient = 16; sizeTemplatex=100; sizeTemplatey=110; GaborScaleList=[ 1.4, 1, 0.7, 0.5]; DoGScaleList =[18.90, 13.36]; LocationShiftLimit=3; OrientShiftLimit=1; #sketches=300;
HMC Parameters: lambdaLearningRate = 0.1/sqrt(sigsq); epsilon = 0.03; L = 10; nIteraton =100; 6x6 chains; sigsq=10;

Case 8: Cougar

Step 1: Sparsification: reconstruction-based filters selection by matching pursuit (arg-max for deformation modeling)

learned templates (Layout of Gabors and DoGs)

deformed templates at different scales

template

 (i) original (ii) templates (iii) reconstruced (iv) residual

Step 2: Simulation: synthesis by learning Frame model (arg-max for deformation modeling)

parameters setting:

General Parameters: nOrient = 16; sizeTemplatex=100; sizeTemplatey=100; GaborScaleList=[ 1.4, 1, 0.7, 0.5]; DoGScaleList =[18.90, 13.36]; LocationShiftLimit=3; OrientShiftLimit=1; #sketches=300;
HMC Parameters: lambdaLearningRate = 0.1/sqrt(sigsq); epsilon = 0.03; L = 10; nIteraton =100; 6x6 chains; sigsq=10;

Case 9: Lion

Step 1: Sparsification: reconstruction-based filters selection by matching pursuit (arg-max for deformation modeling)

learned templates (Layout of Gabors and DoGs)

deformed templates at different scales

template

 (i) original (ii) templates (iii) reconstruced (iv) residual

Step 2: Simulation: synthesis by learning Frame model (arg-max for deformation modeling)

parameters setting:

General Parameters: nOrient = 16; sizeTemplatex=110; sizeTemplatey=110; GaborScaleList=[ 1.4, 1, 0.7, 0.5]; DoGScaleList =[18.90, 13.36]; LocationShiftLimit=3; OrientShiftLimit=1; #sketches=300;
HMC Parameters: lambdaLearningRate = 0.1/sqrt(sigsq); epsilon = 0.03; L = 10; nIteraton =100; 6x6 chains; sigsq=10;