{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/image-classification-and-retrieval-with","title":"Image classification and retrieval with random depthwise signed convolutional neural networks","arxiv_id":"1806.05789","date":"2018-06-15","proceeding":null,"authors":["Yunzhe Xue","Usman Roshan"],"abstract":"We propose a random convolutional neural network to generate a feature space\nin which we study image classification and retrieval performance. Put briefly\nwe apply random convolutional blocks followed by global average pooling to\ngenerate a new feature, and we repeat this k times to produce a k-dimensional\nfeature space. This can be interpreted as partitioning the space of image\npatches with random hyperplanes which we formalize as a random depthwise\nconvolutional neural network. In the network's final layer we perform image\nclassification and retrieval with the linear support vector machine and\nk-nearest neighbor classifiers and study other empirical properties. We show\nthat the ratio of image pixel distribution similarity across classes to within\nclasses is higher in our network's final layer compared to the input space.\nWhen we apply the linear support vector machine for image classification we see\nthat the accuracy is higher than if we were to train just the final layer of\nVGG16, ResNet18, and DenseNet40 with random weights. In the same setting we\ncompare it to an unsupervised feature learning method and find our accuracy to\nbe comparable on CIFAR10 but higher on CIFAR100 and STL10. We see that the\naccuracy is not far behind that of trained networks, particularly in the top-k\nsetting. For example the top-2 accuracy of our network is near 90% on both\nCIFAR10 and a 10-class mini ImageNet, and 85% on STL10. We find that k-nearest\nneighbor gives a comparable precision on the Corel Princeton Image Similarity\nBenchmark than if we were to use the final layer of trained networks. As with\nother networks we find that our network fails to a black box attack even though\nwe lack a gradient and use the sign activation. We highlight sensitivity of our\nnetwork to background as a potential pitfall and an advantage. Overall our work\npushes the boundary of what can be achieved with random weights.","url_abs":"http://arxiv.org/abs/1806.05789v3","url_pdf":"http://arxiv.org/pdf/1806.05789v3.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"image-classification-and-retrieval-with","repo_url":"https://github.com/xyzacademic/RandomDepthwiseCNN","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"classification","task_name":"General Classification"},{"task_slug":"image-classification","task_name":"Image Classification"},{"task_slug":"retrieval","task_name":"Retrieval"},{"task_slug":"image-classification","task_name":"image-classification"}],"methods":[{"method_slug":"average-pooling","method_name":"Average Pooling"},{"method_slug":"global-average-pooling","method_name":"Global Average Pooling"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}