KAKossi Josué Adossehoun

Kossi Josué Adossehoun

*This is an algorithmically generated author record based on publicly available information.

Kossi Josué Adossehoun has a total of 1 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    When and Where to Touch: A Temporal Contact Prior for Text-Driven 3D Hand–Object Interaction

    by Kossi Josué Adossehoun, Hazem Wannous

    Generating realistic 3D hand–object interaction motions from text remains challenging because manipulation requires reasoning about contact as a fundamentally spatio-temporal phenomenon: a model must infer not only where the hands should interactwith the object, but also when such interactions should occur. Existing contact-aware approaches, however, typically represent contact as a static spatial cue or infer it from known object trajectories, limiting their ability to capture the temporal structure specified by language. To address this limitation, we introduce 2WT-HOI, a diffusion-based framework for text-driven hand-object interaction synthesis. At the core of our method is a Temporal Contact Prior (TCP), which predicts, for each hand, frame-wise contact activation and spatial attention over object surface regions from the input text and object features. By explicitly modeling both the timing and location of contact, TCP providesstructured spatio-temporal guidance for motion generation. We integrate this prior into a Transformer-based diffusion model, where it conditions the denoising process toward temporally coherent and spatially plausible interaction patterns while preserving motion diversity. Experiments on multiple HOI benchmarks show that 2WT-HOI generates more realistic and contact-consistent interactions, improves semantic alignment, and substantially reduces penetration artifacts across diverse manipulation scenarios.