Scal3R: 3D generator — what it does and what you need to run it

Scal3R is an open source 3D generator from the ZJU3DV group. The tool tackles the image-to-3D task: you feed it an image, and it outputs a 3D reconstruction. The authors position the project as scalable test-time training for large-scale 3D reconstruction and note that the work was accepted at CVPR 2026 with a Highlight designation.

What it does

As described by the authors, Scal3R is designed for large-scale 3D reconstruction from images. The repository tags highlight three key areas:

• 3D reconstruction;
• depth estimation;
• large-scene handling.

The authors do not provide a detailed list of supported input image formats or output meshes in the fact sheet. Nor do they specify limits on scene size or requirements for the number of viewpoints.

What you need to run it

The code is written in Python. As described by the author, CUDA is required — the repository includes a separate installation guide with instructions for PyTorch and CUDA. Specific library versions and the minimum VRAM amount are not listed in the fact sheet.

The code license is non-standard — the authors do not explicitly state its type, so you need to check the terms manually in the repository before use. The code is open source, and the model is published on Hugging Face.

Who it suits

The tool will appeal to researchers and developers working on 3D scene reconstruction, depth estimation, and those wanting to explore the test-time training approach. The open source code and published model make it possible to understand the method and adapt it to your own data.

For everyday tasks like “upload one photo — get a ready-to-print model,” the project may be overkill: the authors do not describe a ready-made interface, and installation requires working with CUDA and PyTorch. Those not willing to deal with a non-standard license and manual environment setup should look at other solutions.

Scal3R is an open source research tool aimed at large-scale reconstruction. Before using it in your project, it is worth reviewing the installation guide and license terms in the repository.