7ed0ef6193af0bacf9a5ddff2c3d1591 sensors_a2022v22n12p4389.pdf e356eb2979df4a2b3bef05de2fe1be2ccb87956f sensors_a2022v22n12p4389.pdf 0ee4bdb194ede781c53ea77117229a95a761304b40099db6f3391f643510f982 sensors_a2022v22n12p4389.pdf Title: Hierarchical Novelty Detection for Traffic Sign Recognition Subject: Recent works have made significant progress in novelty detection, i.e@汥瑀瑯步渠., the problem of detecting samples of novel classes, never seen during training, while classifying those that belong to known classes. However, the only information this task provides about novel samples is that they are unknown. In this work, we leverage hierarchical taxonomies of classes to provide informative outputs for samples of novel classes. We predict their closest class in the taxonomy, i.e@汥瑀瑯步渠., its parent class. We address this problem, known as hierarchical novelty detection, by proposing a novel loss, namely Hierarchical Cosine Loss that is designed to learn class prototypes along with an embedding of discriminative features consistent with the taxonomy. We apply it to traffic sign recognition, where we predict the parent class semantics for new types of traffic signs. Our model beats state-of-the art approaches on two large scale traffic sign benchmarks, Mapillary Traffic Sign Dataset (MTSD) and Tsinghua-Tencent 100K (TT100K), and performs similarly on natural images benchmarks (AWA2, CUB). For TT100K and MTSD, our approach is able to detect novel samples at the correct nodes of the hierarchy with 81% and 36% of accuracy, respectively, at 80% known class accuracy. Keywords: novelty detection; hierarchical classification; deep learning; traffic sign recognition; autonomous driving; computer vision Author: Idoia Ruiz and Joan Serrat Creator: LaTeX with hyperref Producer: pdfTeX-1.40.21 CreationDate: Mon Jun 13 06:47:55 2022 CEST ModDate: Mon Jun 13 06:52:32 2022 CEST Tagged: no UserProperties: no Suspects: no Form: none JavaScript: no Pages: 22 Encrypted: no Page size: 595.276 x 841.89 pts (A4) Page rot: 0 File size: 2229467 bytes Optimized: no PDF version: 1.7 name type encoding emb sub uni object ID ------------------------------------ ----------------- ---------------- --- --- --- --------- BHNZVG+URWPalladioL-Bold Type 1 Custom yes yes no 10 0 JIUZYB+URWPalladioL-Roma Type 1 Custom yes yes no 15 0 RWQIGL+URWPalladioL-Ital Type 1 Custom yes yes no 19 0 MCBCCI+SourceSansRoman-Regular Type 1C Custom yes yes no 27 0 MCBCCM+SourceSansRoman.613wght Type 1C Custom yes yes no 31 0 MCBCCO+SourceSansRoman-Bold Type 1C Custom yes yes no 35 0 RVSEMT+TimesNewRomanPS-BoldItalicMT Type 1C WinAnsi yes yes no 47 0 GIGFZE+CMR10 Type 1 Builtin yes yes no 84 0 ECNNTI+PazoMath-Italic Type 1 Builtin yes yes no 88 0 QQXUPP+PazoMath Type 1 Builtin yes yes no 92 0 FMDJKA+CMSY10 Type 1 Builtin yes yes no 96 0 RLYTRW+CMMI10 Type 1 Builtin yes yes no 103 0 OKGTOD+URWPalladioL-BoldItal Type 1 Custom yes yes no 138 0 JSTIJA+CMBSY10 Type 1 Builtin yes yes no 142 0 TZDFDH+PazoMath-BoldItalic Type 1 Builtin yes yes no 171 0 BVTOFD+CMBX10 Type 1 Builtin yes yes no 195 0 AAAAAA+LiberationSans CID TrueType Identity-H yes yes yes 207 0 AAAAAA+LiberationSans CID TrueType Identity-H yes yes yes 326 0 AAAAAA+LiberationSans CID TrueType Identity-H yes yes yes 542 0 AAAAAA+LiberationSans CID TrueType Identity-H yes yes yes 608 0 Jhove (Rel. 1.22.1, 2019-04-17) Date: 2023-05-05 02:31:18 CEST RepresentationInformation: sensors_a2022v22n12p4389.pdf ReportingModule: BYTESTREAM, Rel. 1.4 (2018-10-01) LastModified: 2022-06-23 11:39:04 CEST Size: 2229467 Format: bytestream Status: Well-Formed and valid SignatureMatches: PDF-hul MIMEtype: application/octet-stream Checksum: 27ce245a Type: CRC32 Checksum: 7ed0ef6193af0bacf9a5ddff2c3d1591 Type: MD5 Checksum: e356eb2979df4a2b3bef05de2fe1be2ccb87956f Type: SHA-1