Sep 6, 2023 · In this paper, we first propose Vote2Cap-DETR, a simple-yet-effective transformer framework that decouples the decoding process of caption generation and ...

Official implementation of "End-to-End 3D Dense Captioning with Vote2Cap-DETR" (CVPR 2023) and "Vote2Cap-DETR++: Decoupling Localization and Describing for End- ...

(b) The proposed Vote2Cap-DETR frames 3D dense captioning as a set prediction problem and decouple the decoding process of object localization and caption.

We propose two transformer-based 3D dense captioning frameworks that decouple the caption generation from object localization to avoid the cumulative errors ...

3D dense captioning requires a model to translate its understanding of an input 3D scene into several captions associated with different object regions.

In our work, we extend the DETR architecture for 3D dense captioning that makes caption generation and box localization fully in- terrelated with parallel ...

Missing: ++: Decoupling

Apr 16, 2024 · 3D dense captioning requires a model to translate its understanding of an input 3D scene into several captions associated with different object ...

Apr 12, 2024 · 为此，我们提出了一个高级版本Vote2Cap-DETR++，它将查询解耦为本地化和字幕查询，以捕获特定于任务的特征。此外，我们还引入了迭代空间优化策略，对查询进行 ...

Dense captioning in 3D point clouds is an emerging vision-and-language task involving object-level 3D scene understanding.