iFRAME (Inhomogeneous Filters Random Field And Maximum Entropy)

Experiment 3.3: Learning Sparse FRAME Model from Un-aligned Images by Generative Boosting

Code and dataset

Case 1: Maple

Learned models after alignment

(synthesis without local normalization / synthesis with local normalization / wavelet template)

Training images with inferred objects:

aligned images:

parameters setting:

General Parameters: nOrient = 16; GaborScaleList=[1.4, 1, 0.7]; DoGScaleList =[]; LocationShiftLimit=3; OrientShiftLimit=1; interval=1; #Wavelet=250; isLocalNormalize=true
Multiple Selection Parameters: nPartCol=5; nPartRow=5; part_sx=20; part_sy=20; gradient_threshold_scale=0.8
Gibbs Sampler Parameters: lambdaLearningRate = 0.1/sqrt(sigsq); 8x8 chains; sigsq=10; threshold_corrBB=0; lower_bound_rand = 0.001; upper_bound_rand = 0.999; c_val_list=-25:3:25
Alignment Parameters: rotateShiftLimit=16; numResolutionEnlarge = 13; numResolutionShrink = 7; scaleStepSize=0.1; numIterationEM=5; isToUseNewFilterBankAtLastIteration=true; GaborScaleListLastIteration=[1.4, 1, 0.7, 0.5]; DoGScaleListLastIteration=[18.90, 13.36]; isLocalNormalizeLastIteration=false; numWaveletLastIteration = 370;

Case 2: Lion

Learned models after alignment

(synthesis without local normalization / synthesis with local normalization / wavelet template)

Training images with inferred objects:

aligned images:

parameters setting:

General Parameters: nOrient = 16; GaborScaleList=[1.4, 1, 0.7]; DoGScaleList =[]; LocationShiftLimit=3; OrientShiftLimit=1; interval=1; #Wavelet=250; isLocalNormalize=true
Multiple Selection Parameters: nPartCol=5; nPartRow=5; part_sx=20; part_sy=20; gradient_threshold_scale=0.8
Gibbs Sampler Parameters: lambdaLearningRate = 0.1/sqrt(sigsq); 8x8 chains; sigsq=10; threshold_corrBB=0; lower_bound_rand = 0.001; upper_bound_rand = 0.999; c_val_list=-25:3:25
Alignment Parameters: rotateShiftLimit=5; numResolutionEnlarge = 8; numResolutionShrink = 0; scaleStepSize=0.1; numIterationEM=5; isToUseNewFilterBankAtLastIteration=true; GaborScaleListLastIteration=[1.4, 1, 0.7, 0.5]; DoGScaleListLastIteration=[18.90, 13.36]; isLocalNormalizeLastIteration=false; numWaveletLastIteration = 370;

Case 3: Flower

Learned models after alignment

(synthesis without local normalization / synthesis with local normalization / wavelet template)

Training images with inferred objects:

Aligned image:

parameters setting:

General Parameters: nOrient = 16; GaborScaleList=[1.4, 1, 0.7]; DoGScaleList =[]; LocationShiftLimit=3; OrientShiftLimit=1; interval=1; #Wavelet=250; isLocalNormalize=true
Multiple Selection Parameters: nPartCol=5; nPartRow=5; part_sx=20; part_sy=20; gradient_threshold_scale=0.8
Gibbs Sampler Parameters: lambdaLearningRate = 0.1/sqrt(sigsq); 8x8 chains; sigsq=10; threshold_corrBB=0; lower_bound_rand = 0.001; upper_bound_rand = 0.999; c_val_list=-25:3:25
Alignment Parameters: rotateShiftLimit=5; numResolutionEnlarge = 8; numResolutionShrink = 0; scaleStepSize=0.1; numIterationEM=5; isToUseNewFilterBankAtLastIteration=true; GaborScaleListLastIteration=[1.4, 1, 0.7, 0.5]; DoGScaleListLastIteration=[18.90, 13.36]; isLocalNormalizeLastIteration=false; numWaveletLastIteration = 370;

Case 4: Wolf

Learned models after alignment

(synthesis without local normalization / synthesis with local normalization / wavelet template)

Training images with inferred objects:

Aligned image:

parameters setting:

General Parameters: nOrient = 16; GaborScaleList=[1.4, 1, 0.7]; DoGScaleList =[]; LocationShiftLimit=3; OrientShiftLimit=1; interval=1; #Wavelet=250; isLocalNormalize=true
Multiple Selection Parameters: nPartCol=5; nPartRow=5; part_sx=20; part_sy=20; gradient_threshold_scale=0.8
Gibbs Sampler Parameters: lambdaLearningRate = 0.1/sqrt(sigsq); 8x8 chains; sigsq=10; threshold_corrBB=0; lower_bound_rand = 0.001; upper_bound_rand = 0.999; c_val_list=-25:3:25
Alignment Parameters: rotateShiftLimit=5; numResolutionEnlarge = 14; numResolutionShrink = 0; scaleStepSize=0.1; numIterationEM=5; isToUseNewFilterBankAtLastIteration=true; GaborScaleListLastIteration=[1.4, 1, 0.7, 0.5]; DoGScaleListLastIteration=[18.90, 13.36]; isLocalNormalizeLastIteration=false; numWaveletLastIteration = 370;

Case 5: Cougar

Learned models after alignment

(synthesis without local normalization / synthesis with local normalization / wavelet template)

Training images with inferred objects:

Aligned image:

parameters setting:

General Parameters: nOrient = 16; GaborScaleList=[1.4, 1, 0.7]; DoGScaleList =[]; LocationShiftLimit=3; OrientShiftLimit=1; interval=1; #Wavelet=250; isLocalNormalize=true
Multiple Selection Parameters: nPartCol=5; nPartRow=5; part_sx=20; part_sy=20; gradient_threshold_scale=0.8
Gibbs Sampler Parameters: lambdaLearningRate = 0.1/sqrt(sigsq); 8x8 chains; sigsq=10; threshold_corrBB=0; lower_bound_rand = 0.001; upper_bound_rand = 0.999; c_val_list=-25:3:25
Alignment Parameters: rotateShiftLimit=5; numResolutionEnlarge = 14; numResolutionShrink = 0; scaleStepSize=0.1; numIterationEM=5; isToUseNewFilterBankAtLastIteration=true; GaborScaleListLastIteration=[1.4, 1, 0.7, 0.5]; DoGScaleListLastIteration=[18.90, 13.36]; isLocalNormalizeLastIteration=false; numWaveletLastIteration = 370;

Case 6: Hummingbird

Learned models after alignment

(synthesis without local normalization / synthesis with local normalization / wavelet template)

Training images with inferred objects:

Aligned image:

parameters setting:

General Parameters: nOrient = 16; GaborScaleList=[1.4, 1, 0.7]; DoGScaleList =[]; LocationShiftLimit=3; OrientShiftLimit=1; interval=1; #Wavelet=250; isLocalNormalize=true
Multiple Selection Parameters: nPartCol=5; nPartRow=5; part_sx=20; part_sy=20; gradient_threshold_scale=0.8
Gibbs Sampler Parameters: lambdaLearningRate = 0.1/sqrt(sigsq); 8x8 chains; sigsq=10; threshold_corrBB=0; lower_bound_rand = 0.001; upper_bound_rand = 0.999; c_val_list=-25:3:25
Alignment Parameters: rotateShiftLimit=5; numResolutionEnlarge = 10; numResolutionShrink = 0; scaleStepSize=0.1; numIterationEM=5; isToUseNewFilterBankAtLastIteration=true; GaborScaleListLastIteration=[1.4, 1, 0.7, 0.5]; DoGScaleListLastIteration=[18.90, 13.36]; isLocalNormalizeLastIteration=false; numWaveletLastIteration = 370;

Case 7: Goose

Learned models after alignment

(synthesis without local normalization / synthesis with local normalization / wavelet template)

Training images with inferred objects:

Aligned image:

parameters setting:

General Parameters: nOrient = 16; GaborScaleList=[1.4, 1, 0.7]; DoGScaleList =[]; LocationShiftLimit=3; OrientShiftLimit=1; interval=1; #Wavelet=250; isLocalNormalize=true
Multiple Selection Parameters: nPartCol=5; nPartRow=5; part_sx=20; part_sy=20; gradient_threshold_scale=0.8
Gibbs Sampler Parameters: lambdaLearningRate = 0.1/sqrt(sigsq); 8x8 chains; sigsq=10; threshold_corrBB=0; lower_bound_rand = 0.001; upper_bound_rand = 0.999; c_val_list=-25:3:25
Alignment Parameters: rotateShiftLimit=5; numResolutionEnlarge = 6; numResolutionShrink =0; scaleStepSize=0.1; numIterationEM=5; isToUseNewFilterBankAtLastIteration=true; GaborScaleListLastIteration=[1.4, 1, 0.7, 0.5]; DoGScaleListLastIteration=[18.90, 13.36]; isLocalNormalizeLastIteration=false; numWaveletLastIteration = 370;