Automatic discovery and optimization of parts for image classification
Part-based representations have been shown to be very useful for image classification. Learning part-based models is often viewed as a two-stage problem. First, a collection of informative parts is discovered, using heuristics that promote part distinctiveness and diversity, and then classifiers are...
| Main Authors: | , , , |
|---|---|
| Format: | Conference item |
| Language: | English |
| Published: |
Computational and Biological Learning Society
2015
|
| _version_ | 1826315148293308416 |
|---|---|
| author | Parizi, SN Vedaldi, A Felzenszwalb, P Zisserman, A |
| author_facet | Parizi, SN Vedaldi, A Felzenszwalb, P Zisserman, A |
| author_sort | Parizi, SN |
| collection | OXFORD |
| description | Part-based representations have been shown to be very useful for image classification. Learning part-based models is often viewed as a two-stage problem. First, a collection of informative parts is discovered, using heuristics that promote part distinctiveness and diversity, and then classifiers are trained on the vector of part responses. In this paper we unify the two stages and learn the image classifiers and a set of shared parts jointly. We generate an initial pool of parts by randomly sampling part candidates and selecting a good subset using L1/L2 regularization. All steps are driven "directly" by the same objective namely the classification loss on a training set. This lets us do away with engineered heuristics. We also introduce the notion of "negative parts", intended as parts that are negatively correlated with one or more classes. Negative parts are complementary to the parts discovered by other methods, which look only for positive correlations. |
| first_indexed | 2024-12-09T03:20:31Z |
| format | Conference item |
| id | oxford-uuid:9937e9bc-872c-4785-9805-28bbb0a8cbf0 |
| institution | University of Oxford |
| language | English |
| last_indexed | 2024-12-09T03:20:31Z |
| publishDate | 2015 |
| publisher | Computational and Biological Learning Society |
| record_format | dspace |
| spelling | oxford-uuid:9937e9bc-872c-4785-9805-28bbb0a8cbf02024-11-05T14:47:51ZAutomatic discovery and optimization of parts for image classificationConference itemhttp://purl.org/coar/resource_type/c_5794uuid:9937e9bc-872c-4785-9805-28bbb0a8cbf0EnglishSymplectic ElementsComputational and Biological Learning Society2015Parizi, SNVedaldi, AFelzenszwalb, PZisserman, APart-based representations have been shown to be very useful for image classification. Learning part-based models is often viewed as a two-stage problem. First, a collection of informative parts is discovered, using heuristics that promote part distinctiveness and diversity, and then classifiers are trained on the vector of part responses. In this paper we unify the two stages and learn the image classifiers and a set of shared parts jointly. We generate an initial pool of parts by randomly sampling part candidates and selecting a good subset using L1/L2 regularization. All steps are driven "directly" by the same objective namely the classification loss on a training set. This lets us do away with engineered heuristics. We also introduce the notion of "negative parts", intended as parts that are negatively correlated with one or more classes. Negative parts are complementary to the parts discovered by other methods, which look only for positive correlations. |
| spellingShingle | Parizi, SN Vedaldi, A Felzenszwalb, P Zisserman, A Automatic discovery and optimization of parts for image classification |
| title | Automatic discovery and optimization of parts for image classification |
| title_full | Automatic discovery and optimization of parts for image classification |
| title_fullStr | Automatic discovery and optimization of parts for image classification |
| title_full_unstemmed | Automatic discovery and optimization of parts for image classification |
| title_short | Automatic discovery and optimization of parts for image classification |
| title_sort | automatic discovery and optimization of parts for image classification |
| work_keys_str_mv | AT parizisn automaticdiscoveryandoptimizationofpartsforimageclassification AT vedaldia automaticdiscoveryandoptimizationofpartsforimageclassification AT felzenszwalbp automaticdiscoveryandoptimizationofpartsforimageclassification AT zissermana automaticdiscoveryandoptimizationofpartsforimageclassification |