Visual Classification with Multi-Task Joint Sparse Representation |
| Без темы | ||
| << Vision 2050 Generic Presentation | WAAS NOTAM SYSTEM UPDATE >> |
Автор: Markus Svens?n & Christopher M Bishop . Чтобы познакомиться с картинкой полного размера, нажмите на её эскиз. Чтобы можно было использовать все картинки для урока английского языка, скачайте бесплатно презентацию «Visual Classification with Multi-Task Joint Sparse Representation.ppt» со всеми картинками в zip-архиве размером 26333 КБ.
Visual Classification with Multi-Task Joint Sparse Representation
содержание презентации «Visual Classification with Multi-Task Joint Sparse Representation.ppt»| Сл | Текст | Сл | Текст |
| 1 | Visual Classification with Multi-Task | 14 | mapping. Aggregation. Algorithm 2. |
| Joint Sparse Representation. Authors: | MTJSRC-RKHS. | ||
| Xiaotong Yuan and Shuicheng Yan Presenter: | 15 | Column Generation. MTJSRC-CG: take the | |
| Xiaotong Yuan. Learning & Vision | columns of each kernel matrix as feature | ||
| Research Group, ECE, National University | vectors, Objective: Decision: | ||
| of Singapore. | 16 | Experiments. Comparing feature | |
| 2 | Problem to Study. … Task: combine | combination algorithms Nearest Subspace + | |
| multiple features for visual | Combination Sparse Representation + | ||
| classification. Training based methods | Combination Representative kernel feature | ||
| (MKL and SVM ensemble): classifier | combination methods in literature | ||
| training + combination Our solution: cast | (Nilsback, 2008/2009, Varma, 2007, Gehler, | ||
| feature combination to a multi-task joint | 2009). | ||
| sparse representation problem. Color | 17 | Data Sets. Oxford flowers 17 7 Kernels | |
| Texture Shape. In most cases, multiple | (Nilsback, 2009) Oxford flowers 102 4 | ||
| features. | Kernels (Nilsback, 2008) Caltech 101 4 | ||
| 3 | Motivations. … Advances in sparse | Kernels (Varma et al., 2007). | |
| representation (SR) for recognition | 18 | Results on Oxford flowers 17. | |
| (Wright et al., 2009) Robust Training free | Accuracies by Feature Combination. | ||
| Advances in multi-task sparse learning | Accuracies by Single Features. NS. SRC. | ||
| Separate but related sparse learning tasks | MKL (Nilsback, 2009). CG-Boost (Gehler, | ||
| Joint sparsity (Zhang 2006, Liu et al., | 2009). LPBoost (Gehler, 2009). | ||
| 2009). | MTJSRC-RKHS. MTJSRC-CG. 83.2 ± 2.1. 85.9 ± | ||
| 4 | Related Work. Kernel Feature | 2.2. 88.2 ± 2.0. 84.8 ± 2.2. 85.4 ± 2.4. | |
| Combination Multiple Kernel Learning | 88.1 ± 2.3. 88.9 ± 2.9. Features. NS. SVM | ||
| (Varma & Ray, 2007) Boost Individual | (Gehler, 2009). MTJSRC-RKHS (K=1). | ||
| SVM classifier (Gehler & Nowozin, | MTJSRC-CG (K=1). Color. 61.7 ± 3.3. 60.9 ± | ||
| 2009) Multi-task joint covariate selection | 2.1. 64.0 ± 2.1. 64.0 ± 3.3. Shape. 69.9 ± | ||
| (Obozinski et al., 2009) Group Lasso (Yuan | 3.2. 70.2 ± 1.3. 72.7 ± 0.3. 71.5 ± 0.8. | ||
| & Lin, 2006) Multi-task Lasso (Zhang, | Texture. 55.8 ± 1.4. 63.7 ± 2.7. 67.6 ± | ||
| 2006) Sparse representation (SR) for | 2.4. 67.6 ± 2.2. HSV. 61.3 ± 0.7. 62.9 ± | ||
| recognition (Wright et al., 2009). | 2.3. 64.7 ± 4.1. 65.0 ± 3.9. HOG. 57.4 ± | ||
| 5 | Our Method. MTJSRC: a Multi-Task Joint | 3.0. 58.5 ± 4.5. 61.9 ± 3.6. 62.6 ± 2.7. | |
| Sparse Representation and Classification | SIFTint. 70.7 ± 0.7. 70.6 ± 1.6. 74.0 ± | ||
| method Utilizing each feature to form a | 2.2. 74.0 ± 2.0. SIFTbdy. 61.9 ± 4.2. 59.4 | ||
| linear representation task Jointly select | ± 3.3. 62.4 ± 3.2. 63.2 ± 33. | ||
| representative images from few classes | 19 | Results on Oxford flowers 17 (cont.). | |
| Boost the individual features to improve | Sparse representation coefficients and | ||
| performance Extensions in Kernel-view. | reconstruction errors. Color. Shape. | ||
| 6 | Joint Sparse Representation. A set of | Texture. HSV. HOG. SIFTint. SIFTbdy. | |
| images. K types of features. Objective: | 20 | Results on Oxford flowers 102. | |
| block-level sparse coefficients. Test | Accuracies by Feature Combination. | ||
| Image. … … … … … | Accuracies by Single Features. NS. SRC. | ||
| 7 | Formulation. A supervised K-task | MKL (Nilsback, 2008). MTJSRC-RKHS. | |
| linear representation problem. General | MTJSRC-CG. 59.2. 70.0. 72.8. 73.8. 74.1. | ||
| formulation: multi-task least square | Features. NS. SVM (Nilsback, 2008). | ||
| regression with mixed-norm regularization: | MTJSRC-RKHS (K=1). MTJSRC-CG (K=1). HSV. | ||
| Less sparse, convex. Sparser, non-convex. | 39.8. 43.0. 43.6. 42.5. HOG. 34.9. 49.6. | ||
| 8 | Different Mixed-Norms. Joint | 46.7. 48.1. SIFTint. 46.6. 55.1. 54.7. | |
| sparsity-inducing (Obozinski et al., | 55.2. SIFTbdy. 34.1. 32.0. 33.0. 31.6. | ||
| 2009). Joint sparsity-inducing (Zhang, | 21 | Results on Caltech 101. Accuracies by | |
| 2006). K Independent SR tasks. K | Feature Combination (15 train/15 test). | ||
| Independent ridge regression tasks. | Accuracies by Single Features. NS. SRC. | ||
| 9 | Different Mixed-Norms. Joint | MKL (Varma, 2007). LPBoost (Gehler, 2009). | |
| sparsity-inducing (Obozinski et al., | MTJSRC-RKHS. MTJSRC-CG. 51.7 ± 0.8. 69.2 ± | ||
| 2009). Joint sparsity-inducing (Zhang, | 0.7. 70.0 ± 1.0. 70.7 ± 0.4. 71.0 ± 0.3. | ||
| 2006). K Independent SR tasks. K | 71.4 ± 0.4. Features. NS. SVM (Varma, | ||
| Independent ridge regression tasks. | 2007). MTJSRC-RKHS (K=1). MTJSRC-CG (K=1). | ||
| 10 | Optimization. An Accelerated Proximal | GB. 40.8 ± 0.6. 62.6 ± 1.2. 58.3 ± 0.4. | |
| Gradient Method (Tseng, 2008) Generalized | 58.5 ± 0.3. PHOW-gray. 45.4 ± 0.9. 63.9 ± | ||
| gradient mapping step Aggregation step. | 0.8. 65.0 ± 0.7. 64.5 ± 0.5. PHOW-color. | ||
| 11 | Classification. Optimal reconstruction | 37.3 ± 0.5. 54.5 ± 0.6. 56.1 ± 0.5. 54.4 ± | |
| coefficients: Decision: Feature | 0.7. SSIM. 39.8 ± 0.8. 54.3 ± 0.6. 61.8 ± | ||
| confidence: learned via Linear Programming | 0.6. 59.7 ± 0.4. | ||
| Boosting on validation set. | 22 | Conclusions. Multi-task joint sparse | |
| 12 | Algorithm. Algorithm 1. Multi-Task | representation is effective to combine | |
| Joint Sparse Representation | complementary visual features. For single | ||
| Classification. Generalized gradient | feature, the kernel-view extensions of | ||
| mapping. Aggregation. | MTJSRC perform quite competitive to SVM. | ||
| 13 | Kernel-View Extensions. Multitask | MTJSRC is free of model training. | |
| joint sparse representation in a RKHS. The | 23 | Thank you! | |
| APG optimization is characterized by inner | 24 | Visual Classification with Multi-Task | |
| product of feature vectors. training | Joint Sparse Representation. Xiaotong Yuan | ||
| kernel matrix. testing kernel vector. | and Shuicheng Yan. | ||
| 14 | Algorithm. Generalized gradient | ||
| Visual Classification with Multi-Task Joint Sparse Representation.ppt | |||
Visual Classification with Multi-Task Joint Sparse Representation
«Переменные Visual Basic» - Имена переменных. A = 216 b = -31576 c = 3.1415926 D = “visual basic” А = А - 10. Пример программного кода Visual Basic. Присваивание переменным значений. Переменные: тип, имя, значение. Переменная. Byte, short, integer, long, single, double – типы числовых значений. Типы переменных. Объявление переменных.
«Royal family» - Cigar Smoke. The Beatles film. Champagne. Objects of precious metal often worn for personal adornment. Dried leaves of a plant, cultivated in China, India. THE 5TH GAME Traditions and customs. Dictating letters. Milk pudding. Charles Dickens. Quiet evenings at home. THE ROYAL FAMILY. 3. Where was Princess Elizabeth educated? a) at home. b) in Oxford. c) in Cambridge.
«Victory day» - Victory Day (9 May). Never forget the fallen. Population North Korea – 23 million. Whole British Commonwealth – 12 million. A shroud of a sadness still lives in our eyes. Congratulations to the veterans and their ancestors on the Great Victory Day ! World will not survive WW3. Learn the lessons of history.
«Окружающая среда Великобритании» - Зачем стране нужны леса? Тема нашего проекта: «Охрана окружающей среды Великобритании». Разделиться на группы. Возникают вопросы: Как влияет человек на окружающую среду? Что нужно делать, чтобы сохранить планету чистой? Без чего нам нельзя прожить? Окружающая среда. Тестирование Составление отчетной презентации Высказывание на английском языке по данной теме.
«Художественный перевод» - Прогнозирование. Тревога. Творю сам. Мастерство художественного перевода. В свой мир манил. Ожидание. Художественный вкус. Тот, кто увидел вдруг тебя хоть раз, Отворотил свой взор, едва смиряя. Speaks my name. Movie. Переводы учащихся. That voice which calls to me. The phantom of the opera. In sleep he sang to me.
«Английский алфавит для детей» - Идет молва. Собака- dog. Queen- царица. Ключик - key. Буква J. Рентген. Буква S. Буква Y. Пломбир - ice-cream. Буква М. Волк- a wolf. Буква L. Буква V. Буква Е. Игрушка- toy. N висеть не надоест. На охоту вышла С. Буква A. Всадник. Молодой пират. Зонт - umbrella. Дуб - oak-tree. Буква Z. Буква G. Буква F.
