ScriptHOI: Learning Scripted State Transitions for Open-Vocabulary Human-Object Interaction Detection
Minh Anh Nguyen, Quang Huy Tran, Bao Ngoc Le, SuiYang Guang, Tuan Kiet Pham, Linh Chi Vo · 2026
Read ScriptHOI: Learning Scripted State Transitions for Open-Vocabulary Human-Object Interaction Detection by Minh Anh Nguyen, Quang Huy Tran, Bao Ngoc Le, SuiYang Guang, Tuan Kiet Pham, Linh Chi Vo (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-05057.txt.
Description
Minh Anh Nguyen and five co-authors improve open-vocabulary detection of human-object interactions, where the system must recognise action phrases never labelled in training. Vision-language detectors often guess an interaction from objects alone, for instance predicting cutting from a knife and cake being present. Such guesses skip whether hand, contact, and object state truly support the action.
Their ScriptHOI framework writes each phrase as a soft script of state changes, split into slots for body role, contact, shape, affordance, motion, and object state. A visual tokenizer reads the same slots from each detected pair, and a matcher scores coverage and conflict. Those scores calibrate predictions and expose missing visual evidence.
Catalog record
| CREATOR | Minh Anh Nguyen, Quang Huy Tran, Bao Ngoc Le, SuiYang Guang, Tuan Kiet Pham, Linh Chi Vo |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 31 KB |
| LICENSE | http://creativecommons.org/licenses/by-sa/4.0/ |
| RIGHTS | CC BY SA |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.05057