iFRAME (inhomogeneous Filters Random Field And Maximum Entropy)

Experiment 2: Shared Sparse Coding by Learning Sparse FRAME via Generative Boosting

Code and dataset

(1) Wavelet Selection by Generative Boosting (2) Shape Deformation Inferred by local Max (3) Synthesis by Gibbs Sampler on Reconstruction Coefficients (4) Reconstruction by superposition of the selected deformed wavelets

Contents
Case 1 : Cat
Case 2 : Tiger
Case 3 : Wolf
Case 4 : Lion
Case 5 : Bird
Case 6 : Pigeon

Case 1: Cat

Sketch template

Synthesized template

Training images

Deformed templates

Reconstructed images

Residual images


synthesis by learning sparse-Frame model

parameters setting:

General Parameters: nOrient = 16; sizeTemplatex=100; sizeTemplatey=100; GaborScaleList=[ 1.4, 1, 0.7, 0.5]; DoGScaleList =[18.90, 13.36]; LocationShiftLimit=2; OrientShiftLimit=1; interval=3; #Wavelet=800; nIteration= (interval) x (#Wavelet);
Gibbs Sampler Parameters: lambdaLearningRate = 0.1/sqrt(sigsq); 10x10 chains; sigsq=10; threshold_corrBB=0; lower_bound_rand = 0.001; upper_bound_rand = 0.999; c_val_list=-25:2:25

Case 2: Tiger

Sketch template

Synthesized template

Training images

Deformed templates

Reconstructed images

Residual images


synthesis by learning sparse-Frame model

parameters setting:

General Parameters: nOrient = 16; sizeTemplatex=100; sizeTemplatey=100; GaborScaleList=[ 1.4, 1, 0.7, 0.5]; DoGScaleList =[18.90, 13.36]; LocationShiftLimit=2; OrientShiftLimit=1; interval=3; #Wavelet=800; nIteration= (interval) x (#Wavelet);
Gibbs Sampler Parameters: lambdaLearningRate = 0.1/sqrt(sigsq); 10x10 chains; sigsq=10; threshold_corrBB=0; lower_bound_rand = 0.001; upper_bound_rand = 0.999; c_val_list=-25:2:25

Case 3: Wolf

Sketch template

Synthesized template

Training images

Deformed templates

Reconstructed images

Residual images


synthesis by learning sparse-Frame model

parameters setting:

General Parameters: nOrient = 16; sizeTemplatex=100; sizeTemplatey=100; GaborScaleList=[ 1.4, 1, 0.7, 0.5]; DoGScaleList =[18.90, 13.36]; LocationShiftLimit=2; OrientShiftLimit=1; interval=3; #Wavelet=800; nIteration= (interval) x (#Wavelet);
Gibbs Sampler Parameters: lambdaLearningRate = 0.1/sqrt(sigsq); 10x10 chains; sigsq=10; threshold_corrBB=0; lower_bound_rand = 0.001; upper_bound_rand = 0.999; c_val_list=-25:2:25

Case 4: Lion

Sketch template

Synthesized template

Training images

Deformed templates

Reconstructed images

Residual images


synthesis by learning sparse-Frame model

parameters setting:

General Parameters: nOrient = 16; sizeTemplatex=100; sizeTemplatey=100; GaborScaleList=[ 1.4, 1, 0.7, 0.5]; DoGScaleList =[18.90, 13.36]; LocationShiftLimit=2; OrientShiftLimit=1; interval=3; #Wavelet=800; nIteration= (interval) x (#Wavelet);
Gibbs Sampler Parameters: lambdaLearningRate = 0.1/sqrt(sigsq); 10x10 chains; sigsq=10; threshold_corrBB=0; lower_bound_rand = 0.001; upper_bound_rand = 0.999; c_val_list=-25:2:25

Case 5: Bird

Sketch template

Synthesized template

Training images

Deformed templates

Reconstructed images

Residual images


synthesis by learning sparse-Frame model

parameters setting:

General Parameters: nOrient = 16; sizeTemplatex=100; sizeTemplatey=100; GaborScaleList=[ 1.4, 1, 0.7, 0.5]; DoGScaleList =[18.90, 13.36]; LocationShiftLimit=2; OrientShiftLimit=1; interval=3; #Wavelet=800; nIteration= (interval) x (#Wavelet);
Gibbs Sampler Parameters: lambdaLearningRate = 0.1/sqrt(sigsq); 10x10 chains; sigsq=10; threshold_corrBB=0; lower_bound_rand = 0.001; upper_bound_rand = 0.999; c_val_list=-25:2:25

Case 6: Pigeon

Sketch template

Synthesized template

Training images

Deformed templates

Reconstructed images

Residual images


synthesis by learning sparse-Frame model

parameters setting:

General Parameters: nOrient = 16; sizeTemplatex=100; sizeTemplatey=100; GaborScaleList=[ 1.4, 1, 0.7, 0.5]; DoGScaleList =[18.90, 13.36]; LocationShiftLimit=2; OrientShiftLimit=1; interval=3; #Wavelet=800; nIteration= (interval) x (#Wavelet);
Gibbs Sampler Parameters: lambdaLearningRate = 0.1/sqrt(sigsq); 10x10 chains; sigsq=10; threshold_corrBB=0; lower_bound_rand = 0.001; upper_bound_rand = 0.999; c_val_list=-25:2:25