Technical Exchange

<p style="text-align: left;">Paper Title: HoloPart: Generative 3D Part Amodal Segmentation<br><br>Paper URL: https://arxiv.org/abs/2504.07943<br><br>Project Homepage: https://vast-ai-research.github.io/HoloPart<br><br>Code URL: https://github.com/VAST-AI-Research/HoloPart<br><br>Demo: <a href="https://huggingface.co/spaces/VAST-AI/HoloPart" target="_blank" class="ke-link">https://huggingface.co/spaces/VAST-AI/HoloPart</a></p><div class="figure-open" contenteditable="false" data-ke-type="opengraph" data-ke-align="alignCenter" data-og-type="website" data-og-title="HoloPart - a Hugging Face Space by VAST-AI" data-og-description="여기를 눌러 링크를 확인하세요" data-og-host="huggingface.co" data-og-source-url="https://huggingface.co/spaces/VAST-AI/HoloPart" data-og-url="https://huggingface.co/spaces/VAST-AI/HoloPart" data-og-image="http://scrap.kakaocdn.net/dn/O4qgV/bl4QGmJdPuZ/9AZsJjWPHDcUAMWQ6DCqa0/kakaolink40_original.png?width=1200&height=648"><a href="https://huggingface.co/spaces/VAST-AI/HoloPart" target="_blank" data-source-url="https://huggingface.co/spaces/VAST-AI/HoloPart"><div class="og-image"><img class="thumb_img" src="http://scrap.kakaocdn.net/dn/O4qgV/bl4QGmJdPuZ/9AZsJjWPHDcUAMWQ6DCqa0/kakaolink40_original.png?width=1200&height=648" alt="" xxonerror="this.src="//img1.kakaocdn.net/thumb/C200x200/?fname=https%3A%2F%2Ft1.daumcdn.net%2Fcafe_image%2Fcafe_meta_image_190529.png""></div><div class="og-text"><p class="og-title">HoloPart - a Hugging Face Space by VAST-AI</p><p class="og-desc">여기를 눌러 링크를 확인하세요</p><p class="og-host">huggingface.co</p></div></a></div><p style="text-align: left;">github［unirig］:<br>https://github.com/VAST-AI-Research/UniRig<br><br><br><br>I'd like to share a technical article. On GitHub, a team from Tsinghua University in my country has jointly developed a fully automated AI framework model for rigging and skinning weighting. It's open source, so don't worry, although it's not perfect. I previously shared an AI-powered 3D model generator that generates 3D model assets from images. While not perfect, it's usable after some tweaking. I believe the fundamental reason AI-powered production hasn't been commercialized is the rigging and weighting process. Although the process doesn't take long, I believe manual rigging requires a model with a specific pose (either A or T). While AI-powered "white models" are excellent, models with pre-generated poses (not A or T) can be a real headache to rigging. These articles explain how to deploy Unirig locally. Any model can generate skeletons. Since the skeletons can be rigged, adjustments and weighting are all that's needed. This significantly improves production efficiency. However, I'm struggling with dependency library and version compatibility issues, so I'm sorry I can't provide any guidance.<br><br>thank you~</p>