|
Download README.md from zoeloopy/3dgen-multiview: direct link, hf CLI and curl.
- Browser
- Download file 3.22 kB
-
https://huggingface.co/datasets/zoeloopy/3dgen-multiview/resolve/main/README.md
- Command line
-
hf download hf://datasets/zoeloopy/3dgen-multiview/README.md
-
curl -L -o README.md https://huggingface.co/datasets/zoeloopy/3dgen-multiview/resolve/main/README.md
3.22 kB
| license: cc-by-nc-4.0 | |
| pretty_name: 3DGen Multiview | |
| tags: | |
| - image | |
| - video | |
| - 3d | |
| configs: | |
| - config_name: set-01 | |
| data_files: | |
| - split: data | |
| path: data/multiview/set-01/*.tar | |
| - config_name: base-mp4 | |
| data_files: | |
| - split: data | |
| path: data/multiview/base-mp4/*.tar | |
| - config_name: set-06-mp4 | |
| data_files: | |
| - split: data | |
| path: data/multiview/set-06-mp4/*.tar | |
| - config_name: set-07-mp4 | |
| data_files: | |
| - split: data | |
| path: data/multiview/set-07-mp4/*.tar | |
| # 3DGen Multiview | |
| Paired multiview GIFs, PNG images, and text captions from the 3DGen project. Data is released incrementally. The initial release begins with the `set-01` batch. Shards use WebDataset TAR archives and are grouped by source batch. The `data` split does not define a training or evaluation split. | |
| ## Sample format | |
| Each sample has four files sharing the same key. | |
| - `.gif` contains the multiview sequence. | |
| - `.png` contains the paired image. | |
| - `.txt` contains the original caption joined by exact sample ID. | |
| - `.json` contains the source batch and sample ID. | |
| Samples with a missing or invalid media pair or no exact caption match are excluded. Sample IDs that recur across source batches are namespaced by batch. | |
| ## License and attribution | |
| The released dataset is licensed under [Creative Commons Attribution-NonCommercial 4.0 International](https://creativecommons.org/licenses/by-nc/4.0/). Attribute the 3DGen Multiview dataset to the 3DGen project (`zoeloopy/3dgen-multiview`), link the license, and indicate any changes. | |
| ## Download animated GIFs | |
| The dataset viewer renders the `gif` column as a static PNG preview. Saving the preview image downloads that PNG. The original GIF files are stored inside the TAR shards. | |
| [Download an original animated sample](examples/set-01-000000579_seed26.gif?download=true) | |
|  | |
| This verified example has 21 frames at 576 × 576 pixels and lasts 2.1 seconds. | |
| To obtain original GIFs, open **Files and versions**, download a TAR shard from `data/multiview/set-01/`, and extract it. Each sample includes its `.gif`, paired `.png`, caption `.txt`, and metadata `.json`. | |
| For example, download and extract the first shard. | |
| ```bash | |
| hf download zoeloopy/3dgen-multiview \ | |
| data/multiview/set-01/multiview-set-01-00000.tar \ | |
| --repo-type dataset --local-dir ./3dgen-multiview | |
| mkdir -p ./3dgen-multiview/extracted | |
| tar -xf ./3dgen-multiview/data/multiview/set-01/multiview-set-01-00000.tar \ | |
| -C ./3dgen-multiview/extracted | |
| ``` | |
| Open the extracted `.gif` with an animation-capable viewer. In Python, Pillow exposes the frames through `n_frames` and `ImageSequence.Iterator`. | |
| ## MP4 release | |
| The `base-mp4` configuration contains H.264 MP4 videos encoded at CRF 18 in YUV420p, paired with original PNG images and captions. Each sample has `.mp4`, `.png`, `.txt`, and `.json` files with a common key. Metadata records the frame count, duration, and encoding settings. | |
| MP4 encoding is lossy. Conversion checks preserve the source frame count, dimensions, sequence timing, and duration. Original GIFs are retained locally. This configuration begins with an eight-sample pilot shard and is expanded after video playback is verified. | |