iFRAME (inhomogeneous Filters Random Field And Maximum Entropy)

Experiment 3.1: Geometric Transformation of Sparse FRAME (Rotation and flipping)

Code and dataset

The geometrically transformed versions of the learned model can be obtained by directly applying the operations of dilation, rotation, flipping, or even changing the aspect ratio to {B}, without changing the values of lambda. This amounts to simple affine transformations of (x, s, alpha).

Case 1: Cat

Training images

Model learned from training images: sketches template & synthesized image

Rotated versions of Sparse FRAME derived from the above learned model: sketches templates & synthesized images

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; #sketches=300;
HMC Parameters: lambdaLearningRate = 0.1/sqrt(sigsq); epsilon = 0.03; L = 10; nIteraton =100; 6x6 chains; sigsq=10;
Transformation: flipOrNot=0; rotateShiftLimit = 8;

Case 2: Tiger

Training images

Model learned from training images: sketches template & synthesized image

Rotated versions of Sparse FRAME derived from the above learned model: sketches templates & synthesized images

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; #sketches=300;
HMC Parameters: lambdaLearningRate = 0.1/sqrt(sigsq); epsilon = 0.03; L = 10; nIteraton =100; 6x6 chains; sigsq=10;
Transformation: flipOrNot=0; rotateShiftLimit = 8;

Case 3: Lion

Training images

Model learned from training images: sketches template & synthesized image

Rotated versions of Sparse FRAME derived from the above learned model: sketches templates & synthesized images

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; #sketches=300;
HMC Parameters: lambdaLearningRate = 0.1/sqrt(sigsq); epsilon = 0.03; L = 10; nIteraton =100; 6x6 chains; sigsq=10;
Transformation: flipOrNot=0; rotateShiftLimit = 8;

Case 4: Deer

Training images

Model learned from training images: sketches template & synthesized image

Rotated and flipped versions of Sparse FRAME derived from the above learned model: sketches templates & synthesized images

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; #sketches=300;
HMC Parameters: lambdaLearningRate = 0.1/sqrt(sigsq); epsilon = 0.03; L = 10; nIteraton =100; 6x6 chains; sigsq=10;
Transformation: flipOrNot=1; rotateShiftLimit = 8;

Case 5: Hummingbird

Training images

Model learned from training images: sketches template & synthesized image

Rotated and flipped versions of Sparse FRAME derived from the above learned model: sketches templates & synthesized images

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; #sketches=300;
HMC Parameters: lambdaLearningRate = 0.1/sqrt(sigsq); epsilon = 0.03; L = 10; nIteraton =100; 6x6 chains; sigsq=10;
Transformation: flipOrNot=1; rotateShiftLimit = 8;