TL;DR: FlashWorld enables fast (7 seconds on a 1x A100/A800 GPU, 4 seconds on 1x H100/H800 GPU) and high-quality 3D scene generation across diverse scenes, from a single image or text prompt.
LLaVA-3D could perform both 2D and 3D vision-language tasks. The left block (b) shows that compared with previous 3D LMMs, our LLaVA-3D achieves state-of-the-art performance across a wide range of 3D ...
Abstract: Three-dimensional coronary magnetic res- onance angiography (CMRA) requires reconstruction algorithms that can significantly suppress the artifacts encountered in heavily undersampled ...