English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Out-of-distribution (OOD) detection aims to detect "unknown" data whoselabels have not been seen during the in-distribution (ID) training process.Recent progress in representation learning gives rise to distance-based OODdetection that recognizes inputs as ID/OOD according to their relativedistances to the training data of ID classes. Previous approaches calculatepairwise distances relying only on global image representations, which can besub-optimal as the inevitable background clutter and intra-class variation maydrive image-level representations from the same ID class far apart in a givenrepresentation space. In this work, we overcome this challenge by proposingMulti-scale OOD DEtection (MODE), a first framework leveraging both globalvisual information and local region details of images to maximally benefit OODdetection. Specifically, we first find that existing models pretrained byoff-the-shelf cross-entropy or contrastive losses are incompetent to capturevaluable local representations for MODE, due to the scale-discrepancy betweenthe ID training and OOD detection processes. To mitigate this issue andencourage locally discriminative representations in ID training, we proposeAttention-based Local PropAgation (ALPA), a trainable objective that exploits across-attention mechanism to align and highlight the local regions of thetarget objects for pairwise examples. During test-time OOD detection, aCross-Scale Decision (CSD) function is further devised on the mostdiscriminative multi-scale representations to distinguish ID/OOD data morefaithfully. We demonstrate the effectiveness and flexibility of MODE on severalbenchmarks -- on average, MODE outperforms the previous state-of-the-art by upto 19.24% in FPR, 2.77% in AUROC. Code is available athttps://github.com/JimZAI/MODE-OOD.

From Global to Local: Multi-scale Out-of-distribution Detection