| --- |
| license: mit |
| pretty_name: 404-GEN Mini 3D |
| task_categories: |
| - text-to-3d |
| language: |
| - en |
| tags: |
| - synthetic |
| - art |
| size_categories: |
| - 10K<n<100K |
| --- |
| |
| # 404-GEN Mini 3D |
|
|
| This dataset contains over 20,000 3D assets generated with text prompts using 3D Gaussian Splatting, designed for text-to-3D generation tasks. This is a sample of a much larger dataset comprised of 21.5M assets and 40TB in size, available by request at https://dataset.404.xyz |
|
|
| ## Dataset Description |
|
|
| ### Dataset Summary |
|
|
| 404-GEN Mini 3D is a collection of over 20,000 3D assets generated from text prompts on Bittensor Subnet 17, providing mid- and high-quality 3D assets for text-to-3D generation models. The dataset includes .ply.spz files and corresponding .png renders for a variety of objects, creatures, and characters described through natural language. |
|
|
| ### Languages |
|
|
| The dataset uses English for all text prompts. |
|
|
| ## Dataset Structure |
|
|
| ### Data Instances |
|
|
| Each instance in the dataset consists of: |
| - A text prompt describing the 3D object |
| - A compressed .ply in the form of .ply.spz file containing the 3D model data |
| - A PNG render of the 3D model |
|
|
| ### Data Fields |
|
|
| - `prompt`: Text description used to generate the 3D model |
| - `ply`: 3D model file in PLY format (with SPZ compression). For decompression, see Usage Instructions |
| - `png`: 2D render of the 3D model |
|
|
| ### Data Splits |
|
|
| The dataset consists of a single train split. |
|
|
| ## Usage Instructions |
|
|
| ### Decompressing .ply.spz Files |
|
|
| The 3D assets in this dataset are stored in a compressed format (.ply.spz). There are two ways to decompress these files: |
|
|
| 1. **Web Frontend**: We provide a web-based viewer at https://spz.404.xyz where you can upload and view the compressed files directly. |
|
|
| 2. **Local Decompression**: You can decompress the files locally using our code available on GitHub at https://github.com/404-repo/spz |
|
|
| ## Additional Information |
|
|
| ### Licensing Information |
|
|
| This dataset is released under the MIT License. |
|
|
| ### Citation Information |
|
|
| If you use this dataset in your research, please cite: |
|
|
| `@dataset{404_gen_mini_3d,\ |
| author = {404-GEN},\ |
| title = {404-GEN Mini 3D},\ |
| year = {2025},\ |
| publisher = {Hugging Face},\ |
| howpublished = {\url{https://huggingface.co/datasets/404-Gen/404-gen-mini-3d}} |
| }` |
| |
| ### Contributions |
| |
| This dataset was created to support research and development in text-to-3D generation. Contributions to expand or improve the dataset are welcome through pull requests. |
| |
| ## Dataset Creation |
| |
| ### Curation Rationale |
| |
| This dataset was curated to provide diverse examples of text-prompted 3D models for training and evaluating text-to-3D generation systems. |
| |
| ### Source Data |
| |
| #### Initial Data Collection and Normalization |
| |
| The data was collected by generating 3D models from text prompts on Bittensor Subnet 17 using a variety of open-source 3D generative AI models. |
| |
| ### Annotations |
| |
| The dataset does not include additional annotations beyond the original text prompts. |
| |
| ## Considerations for Using the Data |
| |
| ### Social Impact of Dataset |
| |
| This dataset aims to advance the field of text-to-3D generation, enabling more accessible creation of 3D content for various applications including education, design, gaming, and entertainment. |
| |
| ### Discussion of Biases |
| |
| The dataset may reflect biases present in the text-to-3D generation models used to create the assets, as well as potential biases in the selection of prompts. |
| |
| ### Other Known Limitations |
| |
| Due to the decentralized architecture of Bittensor, the network on which the dataset was generated, information about the model and specifications used to generate each asset is unavailable. |
| |
| ## Additional Information |
| |
| ### Dataset Curators |
| |
| 404-GEN |
| |
| ### Maintenance Status |
| |
| No maintenance planned, however a new version may be released when advances are made in open-source 3D generative AI. |
| |
| ### Version and Updates |
| |
| - v1.0.0: Initial release (April 2025) |
| - Complete dataset of 21.5M assets, 40TB of size available by request at https://dataset.404.xyz |
