Supervised learning and detection with sigmoid transformation in Matlab

In experiment 1, we learn the template from different amount of training images with given bounding boxes.

In experiment 2, we compare with the result from Adaboost method.

-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Example 1.1 data, codes, and readme

Experiment 1.1. The 50 training images are 96*160 front posed human. The first block displays the learned active basis consisting of 40 elements. Each element is symbolized by a bar. The rest of the blocks display the observed images and the corresponding deformed active bases. eps

Example 1.2 data, codes, and readme

Experiment 1.2. The 50 training images are 120*180 bicycle. The first block displays the learned active basis consisting of 40 elements. Each element is symbolized by a bar. The rest of the blocks display the observed images and the corresponding deformed active bases. eps

Experiment 1.1 1.2. The ROC of active basis model with sigmoid transformation. The solid curve is obtained from learning 50 bicycle training images. The dashed curve is obtained from learning 50 front posed human training images. eps

Example 1.3 data, codes, and readme

Experiment 1.3. 30 left posed car training images(70*140). Number of elements are 40. eps

Example 1.4 data, codes, and readme

Experiment 1.4. 30 left posed human training images(134*70) . Number of elements are 40. eps

Experiment 1.3 1.4. The ROC of active basis model with sigmoid transformation. The solid curve is obtained from learning 30 right posed car training images. The dashed curve is obtained from learning 30 left posed human training images. eps

Example 1.5 data, codes, and readme

Experiment 1.5. 10 front posed car training images(70*100). Number of elements are 40. eps

Example 1.6 data, codes, and readme

Experiment 1.6. 10 horse training images(120*50). Number of elements are 40. eps

Experiment 1.5 1.6. The ROC of active basis model with sigmoid transformation. The solid curve is obtained from learning 10 front posed car training images. The dashed curve is obtained from learning 10 horse training images. eps

Example 1.7 data, codes, and readme

Experiment 1.7. 5 butterfly training images(100*150). Number of elements are 40. eps

Example 1.8 data, codes, and readme

Experiment 1.8. 5 sailboat training images(150*120). Number of elements are 40. eps

Experiment 1.7 1.8. The ROC of active basis model with sigmoid transformation. The solid curve is obtained from learning 5 butterfy training images. The dashed curve is obtained from learning 5 front sailboat training images. eps

-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Example 2.1 data, codes, and readme

Experiment 2.1. First and second row are the learned templates by active basis model and adaboost method separately. Image size is 100*150. Training image number are 5, 10, and 30.

Experiment 2.1. ROC curves of active basis model and adaboost method. Image are 100*150 butterfly. Training image number are 5, 10, and 30. Iteration 10 times. eps1 eps2 eps3

Example 2.2 data, codes, and readme

Experiment 2.2. First and second row are the learned templates by active basis model and adaboost method separately. Image size is 70*140. Training image number are 10, 20, 30, 50, 70, and 90.

Experiment 2.2. ROC curves of active basis model and adaboost method. Image are 96*160 right posed car. Training image number are 10, 20, 30, 50, 70, and 90. Iteration 10 times. eps1 eps2 eps3 eps4 eps5 eps6

Example 2.3 data, codes, and readme

Experiment 2.3. First and second row are the learned templates by active basis model and adaboost method separately. Image size is 96*160. Training image number are 20, 30, 50, 70, 90, and 110.

Experiment 2.3. ROC curves of active basis model and adaboost method. Image are 96*160 front posed human. Training image number are 20, 30, 50, 70, 90, and 110. Iteration 10 times. eps1 eps2 eps3 eps4 eps5 eps6

Example 2.4 data, codes, and readme

Experiment 2.4. First and second row are the learned templates by active basis model and adaboost method separately. Image size is 120*180. Training image number are 10, 20, and 30.

Experiment 2.4. ROC curves of active basis model and adaboost method. Image are 120*180 bicycle. Training image number are 10, 20, and 30. Iteration 10 times. eps1 eps2

Example 2.5 data, codes, and readme

Experiment 2.5. First and second row are the learned templates by active basis model and adaboost method separately. Image size is 150*120. Training image number are 10, 20, and 30.

Experiment 2.5. ROC curves of active basis model and adaboost method. Image are 150*120 sailboat. Training image number are 10, 20, and 30. Iteration 10 times. eps1 eps2 eps3