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 Tools annual

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 Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

The task of Human-Object Interaction~(HOI) detection could be divided intotwo core problems, i.e., human-object association and interactionunderstanding. In this paper, we reveal and address the disadvantages of theconventional query-driven HOI detectors from the two aspects. For theassociation, previous two-branch methods suffer from complex and costlypost-matching, while single-branch methods ignore the features distinction indifferent tasks. We propose Guided-Embedding Network~(GEN) to attain atwo-branch pipeline without post-matching. In GEN, we design an instancedecoder to detect humans and objects with two independent query sets and aposition Guided Embedding~(p-GE) to mark the human and object in the sameposition as a pair. Besides, we design an interaction decoder to classifyinteractions, where the interaction queries are made of instance GuidedEmbeddings (i-GE) generated from the outputs of each instance decoder layer.For the interaction understanding, previous methods suffer from long-taileddistribution and zero-shot discovery. This paper proposes a Visual-LinguisticKnowledge Transfer (VLKT) training strategy to enhance interactionunderstanding by transferring knowledge from a visual-linguistic pre-trainedmodel CLIP. In specific, we extract text embeddings for all labels with CLIP toinitialize the classifier and adopt a mimic loss to minimize the visual featuredistance between GEN and CLIP. As a result, GEN-VLKT outperforms the state ofthe art by large margins on multiple datasets, e.g., +5.05 mAP on HICO-Det. Thesource codes are available at https://github.com/YueLiao/gen-vlkt.

GEN-VLKT: Simplify Association and Enhance Interaction Understanding for HOI Detection