[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-83402":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":15,"stars7d":13,"stars30d":13,"stars90d":16,"forks30d":16,"starsTrendScore":17,"compositeScore":18,"rankGlobal":10,"rankLanguage":10,"license":10,"archived":19,"fork":19,"defaultBranch":20,"hasWiki":19,"hasPages":21,"topics":22,"createdAt":10,"pushedAt":23,"updatedAt":24,"readmeContent":25,"aiSummary":10,"trendingCount":16,"starSnapshotCount":16,"syncStatus":26,"lastSyncTime":27,"discoverSource":28},83402,"C4G","cvlab-kaist\u002FC4G","cvlab-kaist","Official Implementation of \"Learning Global Motion with Compact Gaussians for Feed-Forward 4D Reconstruction\"","https:\u002F\u002Fcvlab-kaist.github.io\u002FC4G\u002F",null,"Python",55,3,52,1,0,5,42.61,false,"main",true,[],"2026-06-10 03:37:23","2026-06-09 16:35:49","\u003Cp align=\"center\">\n  \u003Ch1 align=\"center\">Learning Global Motion with Compact Gaussians \u003Cbr> for Feed-Forward 4D Reconstruction\u003C\u002Fh1>\n  \u003Cp align=\"center\">\n    \u003Ca href=\"\">Mungyeom Kim\u003C\u002Fa>\u003Csup>1*\u003C\u002Fsup>\n    ·\n    \u003Ca href=\"https:\u002F\u002Fsites.google.com\u002Fview\u002Fminjeon\u002Fhome\">Minkyeong Jeon\u003C\u002Fa>\u003Csup>1*\u003C\u002Fsup>\n    ·\n    \u003Ca href=\"https:\u002F\u002Fhg010303.github.io\u002F\">Honggyu An\u003C\u002Fa>\u003Csup>1*\u003C\u002Fsup>\n    ·\n    \u003Ca href=\"https:\u002F\u002Fcrepejung00.github.io\u002F\">Jaewoo Jung\u003C\u002Fa>\u003Csup>1\u003C\u002Fsup>\n    ·\n    \u003Ca href=\"\">Hyunah Ko\u003C\u002Fa>\u003Csup>1\u003C\u002Fsup>\n    ·\n    \u003Ca href=\"https:\u002F\u002Fonground.github.io\u002F\">Jisang Han\u003C\u002Fa>\u003Csup>1\u003C\u002Fsup>\n    ·\n    \u003Ca href=\"\">Hyeonseo Yu\u003C\u002Fa>\u003Csup>1\u003C\u002Fsup>\n    ·\n    \u003Ca href=\"\">Donghwan Shin\u003C\u002Fa>\u003Csup>1\u003C\u002Fsup>\n    ·\n    \u003Ca href=\"https:\u002F\u002Fsunghwanhong.github.io\u002F\">Sunghwan Hong\u003C\u002Fa>\u003Csup>2\u003C\u002Fsup>\n    ·\n    \u003Ca href=\"\">Takuya Narihira\u003C\u002Fa>\u003Csup>3\u003C\u002Fsup>\n    ·\n    \u003Ca href=\"\">Kazumi Fukuda\u003C\u002Fa>\u003Csup>3\u003C\u002Fsup>\n    ·\n    \u003Ca href=\"https:\u002F\u002Fwww.yukimitsufuji.com\u002F\">Yuki Mitsufuji\u003C\u002Fa>\u003Csup>3,4†\u003C\u002Fsup>\n    ·\n    \u003Ca href=\"https:\u002F\u002Fcvlab.kaist.ac.kr\u002Fmembers\u002Ffaculty\">Seungryong Kim\u003C\u002Fa>\u003Csup>1†\u003C\u002Fsup>\n  \u003C\u002Fp>\n  \u003Ch4 align=\"center\">\u003Csup>1\u003C\u002Fsup>KAIST AI, \u003Csup>2\u003C\u002Fsup>ETH Zurich, ETH AI Center, \u003Csup>3\u003C\u002Fsup>SONY AI, \u003Csup>4\u003C\u002Fsup>Sony Group Corporation\u003C\u002Fh4>\n\n  \u003Cp align='center'>\u003Csup>*\u003C\u002Fsup>Co-first author, †Co-corresponding author\u003C\u002Fp>\n\n  \u003Ch3 align=\"center\">\u003Ca href=\"https:\u002F\u002Fcvlab-kaist.github.io\u002FC4G\">Project Page\u003C\u002Fa>\u003C\u002Fh3>\n  \u003Cdiv align=\"center\">\u003C\u002Fdiv>\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n  \u003Ca href=\"\">\n    \u003Cimg src=\"assets\u002Fteaser.png\" alt=\"C4G teaser\" width=\"90%\">\n  \u003C\u002Fa>\n\u003C\u002Fp>\n\n> We propose a feed-forward dynamic reconstruction network that effectively captures the \u003Cstrong>global motion of dynamic scenes\u003C\u002Fstrong>. Our method employs a timestamp-conditioned, query-based transformer Gaussian decoder that aggregates geometrically consistent features from multi-frame videos, enabling each Gaussian to model globally coherent motion.\n\n### What to Expect\n- [x] Training code for C4G Gaussian reconstruction. \u003Cbr>\n- [x] VDM-based rendering enhancement module code. \u003Cbr>\n- [x] Pretrained C4G weights. \u003Cbr>\n\n## Installation\n\nOur code is developed based on PyTorch 2.2.0, CUDA 12.1, and Python 3.10.\n\nWe recommend using [conda](https:\u002F\u002Fdocs.anaconda.com\u002Fminiconda\u002F) for installation:\n\n```bash\nconda create -n c4g python=3.10\nconda activate c4g\nbash scripts\u002Finstall.sh\n```\n\n## Data Preparation\n\nFor training, we use the preprocessed [RealEstate10K](https:\u002F\u002Fgoogle.github.io\u002Frealestate10k\u002Findex.html) dataset following [pixelSplat](https:\u002F\u002Fgithub.com\u002Fdcharatan\u002Fpixelsplat) and [MVSplat](https:\u002F\u002Fgithub.com\u002Fdonydchen\u002Fmvsplat). Set the dataset path in `config\u002Fdataset\u002Fre10k.yaml` before training.\n\nFor Spring, please refer to the [official website](https:\u002F\u002Fspring-benchmark.org\u002F).\n\nThe default paths are placeholders written as `\u002Fpath\u002Fto\u002F...`.\n\n## Pretrained Weights\n\nC4G initializes from the C3G Gaussian decoder checkpoint. Download `gaussian_decoder.ckpt` from the [C3G Hugging Face repository](https:\u002F\u002Fhuggingface.co\u002FhonggyuAn\u002FC3G\u002Ftree\u002Fmain) and place it under `pretrained_weights\u002Fgaussian_decoder.ckpt`.\n\nC4G checkpoints are available in the [C4G Hugging Face repository](https:\u002F\u002Fhuggingface.co\u002Fmungyeom011\u002FC4G\u002Ftree\u002Fmain).\n\n## Training\n\nTo train C4G, you can run the following commands:\n\n```bash\nbash scripts\u002Ftrain.sh\n```\n\nIf you want to change configs of our training code, you can just modify the main training config in `config\u002Ftraining\u002Fc4g.yaml`.\n\nIf you do not want to log to wandb, keep `wandb.mode=disabled`.\n\n## VDM-based Rendering Enhancement Module\n\nThe optional VDM-based rendering enhancement module code is included under [submodules\u002FDiffSynth-Studio](submodules\u002FDiffSynth-Studio).\n\nTo train and inference the VDM-based rendering enhancement module, please follow [submodules\u002FDiffSynth-Studio\u002FREADME.md](submodules\u002FDiffSynth-Studio\u002FREADME.md).\n\n## Citation\n\n```bibtex\n@article{kim2026learning,\n  title={Learning Global Motion with Compact Gaussians for Feed-Forward 4D Reconstruction},\n  author={Kim, Mungyeom and Jeon, Minkyeong and An, Honggyu and Jung, Jaewoo and Ko, Hyuna and Han, Jisang and Yu, Hyeonseo and Shin, Donghwan and Hong, Sunghwan and Narihira, Takuya and others},\n  journal={arXiv preprint arXiv:2605.31595},\n  year={2026}\n}\n```\n\n## Acknowledgement\n\nWe thank the authors of [VGGT](https:\u002F\u002Fgithub.com\u002Ffacebookresearch\u002Fvggt), [MoGe](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FMoGe), and [CoWTracker](https:\u002F\u002Fgithub.com\u002Ffacebookresearch\u002Fcowtracker) for their excellent work and code.\n",2,"2026-06-11 04:11:05","CREATED_QUERY"]