Practical overview
When it fits
Use GCD when you need a research codebase for dynamic-scene novel-view synthesis and are prepared to install its Python and GPU-oriented dependencies. For casual custom-video inference, follow the Gradio instructions; for quantitative evaluation or saved command-line outputs, use the separate evaluation workflow described by the repository.
Who it is for
- Computer-vision and graphics researchers studying dynamic novel-view synthesis.
- Developers who want to explore a research workflow for generating camera views from monocular dynamic-scene footage.
Documented uses
- Generate novel views of a dynamic scene from monocular video using a pretrained model.
- Explore camera-motion controls through the documented Gradio inference example.
- Train or evaluate models using the project’s dataset workflows.
- Generate datasets with the repository’s dataset-generation tools.
Requirements & limits
The public GitHub repository is open source under GPL-3.0. The documented setup uses Linux-oriented CUDA/PyTorch commands; no hosted service is described.
- The documented environment uses Python 3.10, PyTorch and CUDA 12.1, plus CLIP, Stability AI datapipelines and requirements.txt dependencies.
- The listed pretrained checkpoints are 20.3 GB each, making storage a meaningful requirement.
- Checkpoint configurations differ by dataset, output modality and camera-motion setting; choose the matching config and model path for the intended task.
- The README distinguishes casual Gradio inference from quantitative evaluation and command-line output workflows.
- The repository author says the public code had not been thoroughly vetted.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Set up the documented GCD environment
Create the project’s Python environment and install its listed dependencies.
Before you start
- Conda
- Python 3.10 environment
- CUDA-capable PyTorch installation using the documented CUDA 12.1 package
- Repository clone containing requirements.txt
Steps
Create and activate a Conda environment named gcd with Python 3.10.
Install PyTorch, torchvision and torchaudio with pytorch-cuda=12.1 from the PyTorch and NVIDIA channels.
Install OpenAI CLIP and Stability AI datapipelines from their Git repositories.
Install packages listed in the repository’s requirements.txt.
Expected outcome
The documented Python environment and project dependencies are installed; the repository’s inference instructions provide the next stage.
Guide limitations
- The README does not pin package versions in requirements.txt and says the code has not been thoroughly vetted.
- This setup guide does not establish that inference will run successfully on a particular machine.
Launch the Kubric-4D Gradio inference example
Start the documented Gradio interface with the Kubric gradual max-90-degree checkpoint.
Before you start
- The repository and its documented Python dependencies are set up.
- A Kubric checkpoint named kubric_gradual_max90.ckpt is present in pretrained/.
- A CUDA device is available for the example command.
Steps
From the repository, change into gcd-model/.
Run scripts/gradio_app.py with CUDA_VISIBLE_DEVICES=0, port 7880, configs/infer_kubric.yaml and the pretrained Kubric checkpoint path.
Use the documented output_path, examples_path and task_desc arguments for the Kubric-4D example.
Expected outcome
The command starts the project’s Gradio inference application configured for Kubric-4D dynamic view synthesis up to 90 degrees azimuth.
Guide limitations
- The README lists the checkpoint at 20.3 GB; it must be available at the specified path.
- This is the documented casual inference route, not the repository’s quantitative evaluation workflow.
- The instructions do not guarantee successful operation on a given system.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 17 October 2026
- Original website title
- basilevh/gcd · Official homepage title · retained in its source language