Classifier-as-a-Service: Online Query of Cascades and Operating Points

We introduce a classifier and parameter selection algorithm for Classifier-as-a-Service applications where there are many components (e.g., features, kernels, classifiers) available to construct classification algorithms. Queries specify varying requirements (i.e., quality and execution time), some of which may require combining classification algorithms to satisfy; each query may have a different set of quality and execution time requirements (e.g., fast and precise, slow and thorough) and the set of images to which the classifier is to be applied may be small (e.g., even a single image), necessitating a query resolution method that takes negligible time in comparison. When operating on large datasets, meeting design requirements automatically becomes essential to reducing costs associated with unnecessary computation and expert assistance. As queries specify requirements and not implementation details, additional components can be utilized naturally. Our query resolution method combines classifiers with complementary operating points (e.g., high recall algorithmic filter, followed by high precision human verification) in a rejection-chain configuration. Experiments are conducted on the SUN397[1] dataset; we achieve state-of-the-art classification results and 1 m.s. query resolution times.