Improving Human-Object Interaction Detection via Phrase Learning and Label Composition

Human-Object Interaction (HOI) detection is a fundamental task in high-level human-centric scene understanding. We propose PhraseHOI, containing HOI branch and novel phrase branch, to leverage language prior improve relation expression. Specifically, the supervised by semantic embeddings, whose ground truths are automatically converted from original annotations without extra human efforts. Meanwhile, label composition method proposed deal with long-tailed problem HOI, which composites labels neighbors. Further, optimize loss composed of distilling balanced triplet proposed. Extensive experiments conducted prove effectiveness achieves significant improvement over baseline surpasses previous state-of-the-art methods on Full NonRare challenging HICO-DET benchmark.