CARI4D: Category Agnostic 4D Reconstruction of Human-Object Interaction
CARI4D reconstructs 4D human-object interactions from single RGB videos at metric scale without needing pre-defined object models. It achieves this by combining coarse-to-fine shape estimation, foundation model pose initialization, and a category-agnostic contact reasoning model (CoCoNet). The method significantly outperforms prior work in reconstruction accuracy and generalizes well to unseen objects.