{"id":18427692,"url":"https://github.com/eldar/flash3d","last_synced_at":"2025-10-25T02:28:20.190Z","repository":{"id":249261203,"uuid":"831011683","full_name":"eldar/flash3d","owner":"eldar","description":"Flash3D: Feed-Forward Generalisable 3D Scene Reconstruction from a Single Image","archived":false,"fork":false,"pushed_at":"2024-11-27T15:13:13.000Z","size":14471,"stargazers_count":213,"open_issues_count":15,"forks_count":17,"subscribers_count":6,"default_branch":"main","last_synced_at":"2025-06-01T01:53:13.228Z","etag":null,"topics":[],"latest_commit_sha":null,"homepage":"","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/eldar.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":"AUTHORS","dei":null,"publiccode":null,"codemeta":null,"zenodo":null}},"created_at":"2024-07-19T13:06:25.000Z","updated_at":"2025-05-31T22:26:51.000Z","dependencies_parsed_at":"2025-05-17T14:18:42.640Z","dependency_job_id":null,"html_url":"https://github.com/eldar/flash3d","commit_stats":null,"previous_names":["eldar/flash3d"],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/eldar/flash3d","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/eldar%2Fflash3d","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/eldar%2Fflash3d/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/eldar%2Fflash3d/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/eldar%2Fflash3d/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/eldar","download_url":"https://codeload.github.com/eldar/flash3d/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/eldar%2Fflash3d/sbom","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":266076096,"owners_count":23872731,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":[],"created_at":"2024-11-06T05:11:31.774Z","updated_at":"2025-10-25T02:28:20.099Z","avatar_url":"https://github.com/eldar.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"[![arXiv](https://img.shields.io/badge/arXiv-2406.04343-blue?logo=arxiv\u0026color=%23B31B1B)](https://arxiv.org/abs/2406.04343)\n[![ProjectPage](https://img.shields.io/badge/Project_Page-Flash3D-blue)](https://www.robots.ox.ac.uk/~vgg/research/flash3d/)\n[![HuggingFace](https://img.shields.io/badge/%F0%9F%A4%97%20HuggingFace-Demo-yellow)](https://huggingface.co/spaces/szymanowiczs/flash3d) \n\n\n# Flash3D: Feed-Forward Generalisable 3D Scene Reconstruction from a Single Image\n\n\n\u003cp align=\"center\"\u003e\n  \u003cimg src=\"assets/teaser_video.gif\" alt=\"animated\" /\u003e\n\u003c/p\u003e\n\n\u003e [Flash3D: Feed-Forward Generalisable 3D Scene Reconstruction from a Single Image](https://www.robots.ox.ac.uk/~vgg/research/flash3d/)  \n\u003e Stanislaw Szymanowicz, Eldar Insafutdinov, Chuanxia Zheng, Dylan Campbell, João F. Henriques, Christian Rupprecht, Andrea Vedaldi  \n\u003e 3DV, 2025.\n\u003e *[arXiv 2406.04343](https://arxiv.org/pdf/2406.04343.pdf)*  \n\n# News\n- [x] `19.07.2024`: Training code and data release\n\n# Setup\n\n## Create a python environment\n\nFlash3D has been trained and tested with the followings software versions:\n\n- Python 3.10\n- Pytorch 2.2.2\n- CUDA 11.8\n- GCC 11.2 (or more recent)\n\nBegin by installing CUDA 11.8 and adding the path containing the `nvcc` compiler to the `PATH` environmental variable.\nThen the python environment can be created either via conda:\n\n```sh\nconda create -y python=3.10 -n flash3d\nconda activate flash3d\n```\n\nor using Python's venv module (assuming you already have access to Python 3.10 on your system):\n\n```sh\npython3.10 -m venv .venv\n. .venv/bin/activate\n```\n\nFinally, install the required packages as follows:\n\n```sh\npip install -r requirements-torch.txt --extra-index-url https://download.pytorch.org/whl/cu118\npip install -r requirements.txt\n```\n\n## Download training data\n\n### RealEstate10K dataset\n\nFor downloading the RealEstate10K dataset we base our instructions on the [Behind The Scenes](https://github.com/Brummi/BehindTheScenes/tree/main?tab=readme-ov-file#-datasets) scripts.\nFirst you need to download the video sequence metadata including camera poses from https://google.github.io/realestate10k/download.html and unpack it into `data/` such that the folder layout is as follows:\n\n```\ndata/RealEstate10K/train\ndata/RealEstate10K/test\n```\n\nFinally download the training and test sets of the dataset with the following commands:\n\n```sh\npython datasets/download_realestate10k.py -d data/RealEstate10K -o data/RealEstate10K -m train\npython datasets/download_realestate10k.py -d data/RealEstate10K -o data/RealEstate10K -m test\n```\n\nThis step will take several days to complete. Finally, download additional data for the RealEstate10K dataset.\nIn particular, we provide pre-processed COLMAP cache containing sparse point clouds which are used to estimate the scaling factor for depth predictions.\nThe last two commands filter the training and testing set from any missing video sequences.\n\n```sh\nsh datasets/dowload_realestate10k_colmap.sh\npython -m datasets.preprocess_realestate10k -d data/RealEstate10K -s train\npython -m datasets.preprocess_realestate10k -d data/RealEstate10K -s test\n```\n\n## Download and evaluate the pretrained model\n\nWe provide model weights that could be downloaded and evaluated on RealEstate10K test set:\n\n```sh\npython -m misc.download_pretrained_models -o exp/re10k_v2\nsh evaluate.sh exp/re10k_v2\n```\n\n## Training\n\nIn order to train the model on RealEstate10K dataset execute this command:\n```sh\npython train.py \\\n  +experiment=layered_re10k \\\n  model.depth.version=v1 \\\n  train.logging=false \n```\n\nFor multiple GPU, we can run with this command:\n```sh\nsh train.sh\n```\nYou can modify the cluster information in ```configs/hydra/cluster```.\n\n\n## BibTeX\n```\n@article{szymanowicz2024flash3d,\n      author = {Szymanowicz, Stanislaw and Insafutdinov, Eldar and Zheng, Chuanxia and Campbell, Dylan and Henriques, Joao and Rupprecht, Christian and Vedaldi, Andrea},\n      title = {Flash3D: Feed-Forward Generalisable 3D Scene Reconstruction from a Single Image},\n      journal = {arxiv},\n      year = {2024},\n}\n```\n\n\n\n\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Feldar%2Fflash3d","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Feldar%2Fflash3d","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Feldar%2Fflash3d/lists"}