2020 · 132 citations · 47 references
ComputingComputer ScienceComputer VisionMachine LearningNatural Language ProcessingDeep LearningInstance SegmentationObject DetectionObject RecognitionRelation ExtractionCascade NetworksFine-grained RecognitionInstance LocalizationInteraction RecognitionRelation Prediction
Rapid progress has been witnessed for human‑object interaction (HOI) recognition, but most existing models are confined to single‑stage reasoning pipelines. The study introduces a cascade architecture for multi‑stage, coarse‑to‑fine HOI understanding to address the task’s intrinsic complexity. The method employs a cascade of instance localization and interaction recognition networks that refine HOI proposals across stages, with cross‑stage connections, a relation ranking module, a triple‑stream classifier, and human‑centric relation features, enabling both bounding‑box relation detection and fine‑grained pixel‑wise segmentation. The approach achieved first place in the ICCV2019 Person in Context Challenge for both relation detection and segmentation, and showed promising results on V‑COCO.
Rapid progress has been witnessed for human-object interaction (HOI) recognition, but most existing models are confined to single-stage reasoning pipelines. Considering the intrinsic complexity of the task, we introduce a cascade architecture for a multi-stage, coarse-to-fine HOI understanding. At each stage, an instance localization network progressively refines HOI proposals and feeds them into an interaction recognition network. Each of the two networks is also connected to its predecessor at the previous stage, enabling cross-stage information propagation. The interaction recognition network has two crucial parts: a relation ranking module for high-quality HOI proposal selection and a triple-stream classifier for relation prediction. With our carefully-designed human-centric relation features, these two modules work collaboratively towards effective interaction understanding. Further beyond relation detection on a bounding-box level, we make ourframework flexible to perform fine-grained pixel-wise relation segmentation; this provides a new glimpse into better relation modeling. Our approach reached the 1st place in the ICCV2019 Person in Context Challenge, on both relation detection and segmentation tasks. It also shows promising results on V-COCO.
47
Deep Residual Learning for Image Recognition
Kaiming He, Xiangyu Zhang, Shaoqing Ren et al. · 2016 · 222.1K citations · Full text
Optimization by Simulated Annealing
Scott Kirkpatrick, C. D. Gelatt, M.P. Vecchi · Science · 1983 · 44.6K citations
Kaiming He, Georgia Gkioxari, Piotr Dollár et al. · 2017 · 28.8K citations