ShapeR-Evaluation

# ShapeR Evaluation Dataset  We introduce a new dataset of in-the-wild sequences with paired posed multi-view images, SLAM point clouds, and individually complete 3D shape annotations for 178 objects across 7 diverse scenes. In contrast to existing real-world 3D reconstruction datasets which are either captured in controlled setups or have merged object and background geometries or incomplete shapes, this dataset is designed to capture real-world challenges like occlusions, clutter, and variable resolution and viewpoints to enable realistic, in-the-wild evaluation. [Project Page](http://facebookresearch.github.io/ShapeR) | [Paper](https://huggingface.co/papers/2601.11514) | [Code](https://github.com/facebookresearch/ShapeR) | [Video](https://www.youtube.com/watch?v=EbY30KAA55I) | [HF-Model](https://huggingface.co/facebook/ShapeR/) | [HF Evaluation Dataset](https://huggingface.co/datasets/facebook/ShapeR-Evaluation) ## Usage Clone the [repository](https://github.com/facebookresearch/ShapeR) and follow the INSTALL.md instructions to install the required dependencies. To run inference on a sample from the dataset: ```bash python infer_shape.py --input_pkl <sample.pkl> --config balance --output_dir output ``` ## Examples  The dataset contains 178 objects across seven casually-captured recordings from distinct cluttered scenes annotated high quality 3D geometry. It covers a wide range of categories, from large objects like furniture to smaller items such as remotes, toasters, and tools. For each sequence, we provide multi-view images, calibrated camera parameters, SLAM point clouds, and machine-generated object captions. Each annotated object also includes a complete reference mesh generated using internal image-to-3D modeling methods under ideal conditions, which we manually refined and realigned for geometric and pose consistency. ## Data Format [**ShapeR Evaluation Dataset**](https://huggingface.co/datasets/facebook/ShapeR-Evaluation) contains preprocessed samples from Aria glasses captures, where each sample is a pickle file with point clouds, multi-view images, camera parameters, text captions, and ground truth meshes. For a detailed walkthrough of the data format, see the **[`explore_data.ipynb`](https://github.com/facebookresearch/ShapeR/blob/main/explore_data.ipynb)** notebook which includes: - Complete pickle file structure with all keys and their dimensions - Interactive 3D visualization of point clouds and meshes - Camera position visualization - Image and mask grid displays - DataLoader usage examples for both SLAM and RGB variants - Explanation of view selection strategies ## License ShapeR evaluation dataset is licensed under CC-BY-NC. See [LICENSE](LICENSE) for details. ## Citation If you find ShapeR useful for your research, please cite our paper: ```bibtex @misc{siddiqui2026shaperrobustconditional3d, title={ShapeR: Robust Conditional 3D Shape Generation from Casual Captures}, author={Yawar Siddiqui and Duncan Frost and Samir Aroudj and Armen Avetisyan and Henry Howard-Jenkins and Daniel DeTone and Pierre Moulon and Qirui Wu and Zhengqin Li and Julian Straub and Richard Newcombe and Jakob Engel}, year={2026}, eprint={2601.11514}, archivePrefix={arXiv}, primaryClass={cs.CV}, url={https://arxiv.org/abs/2601.11514}, } ```