[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-72592":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":17,"stars7d":17,"stars30d":18,"stars90d":16,"forks30d":16,"starsTrendScore":19,"compositeScore":20,"rankGlobal":10,"rankLanguage":10,"license":21,"archived":22,"fork":22,"defaultBranch":23,"hasWiki":22,"hasPages":22,"topics":24,"createdAt":10,"pushedAt":10,"updatedAt":28,"readmeContent":29,"aiSummary":30,"trendingCount":16,"starSnapshotCount":16,"syncStatus":17,"lastSyncTime":31,"discoverSource":32},72592,"LGM","3DTopia\u002FLGM","3DTopia","[ECCV 2024 Oral] LGM: Large Multi-View Gaussian Model for High-Resolution 3D Content Creation.","https:\u002F\u002Fme.kiui.moe\u002Flgm\u002F",null,"Python",2085,139,30,60,0,2,12,6,64.64,"MIT License",false,"main",[25,26,27],"gaussian-splatting","image-to-3d","text-to-3d","2026-06-12 04:01:06","\n## Large Multi-View Gaussian Model\n\nThis is the official implementation of *LGM: Large Multi-View Gaussian Model for High-Resolution 3D Content Creation*.\n\n### [Project Page](https:\u002F\u002Fme.kiui.moe\u002Flgm\u002F) | [Arxiv](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.05054) | [Weights](https:\u002F\u002Fhuggingface.co\u002Fashawkey\u002FLGM) | \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Fspaces\u002Fashawkey\u002FLGM\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F%F0%9F%A4%97%20Gradio%20Demo-Huggingface-orange\">\u003C\u002Fa>\n\nhttps:\u002F\u002Fgithub.com\u002F3DTopia\u002FLGM\u002Fassets\u002F25863658\u002Fcf64e489-29f3-4935-adba-e393a24c26e8\n\n### News\n[2024.4.3] Thanks to [@yxymessi](https:\u002F\u002Fgithub.com\u002Fyxymessi) and [@florinshen](https:\u002F\u002Fgithub.com\u002Fflorinshen), we have fixed a **severe bug in rotation normalization** [here](https:\u002F\u002Fgithub.com\u002F3DTopia\u002FLGM\u002Fcommit\u002F9a0797cdbacf8e6216d0108cb00cbe43b9cb3d81). We have finetuned the model with correct normalization for 30 more epochs and uploaded new checkpoints.\n\n### Replicate Demo:\n* gaussians: [demo](https:\u002F\u002Freplicate.com\u002Fcamenduru\u002Flgm) | [code](https:\u002F\u002Fgithub.com\u002Fcamenduru\u002FLGM-replicate)\n* mesh: [demo](https:\u002F\u002Freplicate.com\u002Fcamenduru\u002Flgm-ply-to-glb) | [code](https:\u002F\u002Fgithub.com\u002Fcamenduru\u002FLGM-ply-to-glb-replicate)\n\nThanks to [@camenduru](https:\u002F\u002Fgithub.com\u002Fcamenduru)!\n\n### Install\n\n```bash\n# xformers is required! please refer to https:\u002F\u002Fgithub.com\u002Ffacebookresearch\u002Fxformers for details.\n# for example, we use torch 2.1.0 + cuda 11.8\npip install torch==2.1.0 torchvision==0.16.0 torchaudio==2.1.0 --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fcu118\npip install -U xformers --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fcu118\n\n# a modified gaussian splatting (+ depth, alpha rendering)\ngit clone --recursive https:\u002F\u002Fgithub.com\u002Fashawkey\u002Fdiff-gaussian-rasterization\npip install .\u002Fdiff-gaussian-rasterization\n\n# for mesh extraction\npip install git+https:\u002F\u002Fgithub.com\u002FNVlabs\u002Fnvdiffrast\n\n# other dependencies\npip install -r requirements.txt\n```\n\n### Pretrained Weights\n\nOur pretrained weight can be downloaded from [huggingface](https:\u002F\u002Fhuggingface.co\u002Fashawkey\u002FLGM).\n\nFor example, to download the fp16 model for inference:\n```bash\nmkdir pretrained && cd pretrained\nwget https:\u002F\u002Fhuggingface.co\u002Fashawkey\u002FLGM\u002Fresolve\u002Fmain\u002Fmodel_fp16_fixrot.safetensors\ncd ..\n```\n\nFor [MVDream](https:\u002F\u002Fgithub.com\u002Fbytedance\u002FMVDream) and [ImageDream](https:\u002F\u002Fgithub.com\u002Fbytedance\u002FImageDream), we use a [diffusers implementation](https:\u002F\u002Fgithub.com\u002Fashawkey\u002Fmvdream_diffusers).\nTheir weights will be downloaded automatically.\n\n### Inference\n\nInference takes about 10GB GPU memory (loading all imagedream, mvdream, and our LGM).\n\n```bash\n### gradio app for both text\u002Fimage to 3D\npython app.py big --resume pretrained\u002Fmodel_fp16.safetensors\n\n### test\n# --workspace: folder to save output (*.ply and *.mp4)\n# --test_path: path to a folder containing images, or a single image\npython infer.py big --resume pretrained\u002Fmodel_fp16.safetensors --workspace workspace_test --test_path data_test \n\n### local gui to visualize saved ply\npython gui.py big --output_size 800 --test_path workspace_test\u002Fsaved.ply\n\n### mesh conversion\npython convert.py big --test_path workspace_test\u002Fsaved.ply\n```\n\nFor more options, please check [options](.\u002Fcore\u002Foptions.py).\n\n### Training\n\n**NOTE**: \nSince the dataset used in our training is based on AWS, it cannot be directly used for training in a new environment.\nWe provide the necessary training code framework, please check and modify the [dataset](.\u002Fcore\u002Fprovider_objaverse.py) implementation!\n\nWe also provide the **~80K subset of [Objaverse](https:\u002F\u002Fobjaverse.allenai.org\u002Fobjaverse-1.0)** used to train LGM in [objaverse_filter](https:\u002F\u002Fgithub.com\u002Fashawkey\u002Fobjaverse_filter).\n\n```bash\n# debug training\naccelerate launch --config_file acc_configs\u002Fgpu1.yaml main.py big --workspace workspace_debug\n\n# training (use slurm for multi-nodes training)\naccelerate launch --config_file acc_configs\u002Fgpu8.yaml main.py big --workspace workspace\n```\n\n### Acknowledgement\n\nThis work is built on many amazing research works and open-source projects, thanks a lot to all the authors for sharing!\n\n- [gaussian-splatting](https:\u002F\u002Fgithub.com\u002Fgraphdeco-inria\u002Fgaussian-splatting) and [diff-gaussian-rasterization](https:\u002F\u002Fgithub.com\u002Fgraphdeco-inria\u002Fdiff-gaussian-rasterization)\n- [nvdiffrast](https:\u002F\u002Fgithub.com\u002FNVlabs\u002Fnvdiffrast)\n- [dearpygui](https:\u002F\u002Fgithub.com\u002Fhoffstadt\u002FDearPyGui)\n- [tyro](https:\u002F\u002Fgithub.com\u002Fbrentyi\u002Ftyro)\n\n### Citation\n\n```\n@article{tang2024lgm,\n  title={LGM: Large Multi-View Gaussian Model for High-Resolution 3D Content Creation},\n  author={Tang, Jiaxiang and Chen, Zhaoxi and Chen, Xiaokang and Wang, Tengfei and Zeng, Gang and Liu, Ziwei},\n  journal={arXiv preprint arXiv:2402.05054},\n  year={2024}\n}\n```\n","LGM是一个用于高分辨率3D内容创建的大型多视角高斯模型。该项目的核心功能是通过图像或文本生成高质量的3D模型，基于改进的高斯点云渲染技术，并支持深度和透明度渲染。它特别适用于需要从多个视角图像中重建精细3D场景的应用场景，如虚拟现实、增强现实及游戏开发等。项目提供了预训练权重、详细的安装指南以及多种演示示例，便于用户快速上手体验其强大功能。采用Python语言编写，开放源代码遵循MIT许可证，鼓励社区贡献与协作。","2026-06-11 03:42:42","high_star"]