--- tags: - unified multimodal model - camera-centric - generation - understanding - spatial intelligence - 3D vision task_categories: - text-to-image - image-to-text - image-to-3d - image-to-image pretty_name: Puffin-16M size_categories: - 10M πŸ“– Project Page   |   πŸ’» GitHub   |   πŸ€— Models   |   πŸ“„ HF Paper   |   πŸ€— HF Blog

## Dataset Details Datasets and benchmarks that span vision, language, and camera modalities remain scarce in the domain of spatial multimodal intelligence. **Puffin-16M** is a large-scale, camera-centric dataset that substantially scales up [Puffin-4M](https://huggingface.co/datasets/KangLiao/Puffin-4M), comprising **~16.5 million samples** across two complementary parts: - **Puffin-Cam-15M** β€” **15,338,221** single-view image–caption–camera triplets. Each perspective image (rendered from panoramas across diverse indoor and outdoor scenes) is paired with a caption that combines a scene description, precise camera parameters (roll, pitch, field-of-view, and radial distortion), and a camera-height category. - **Puffin-Traj-1M** β€” **1,136,696** camera trajectories. Each trajectory is a 90-frame sequence with dense per-frame camera annotations (Euler angles, intrinsics, and camera pose), supporting camera-controlled world exploration and trajectory-aware generation and understanding. Two small **held-out benchmarks** accompany the training corpus for evaluation, rendered from panoramas disjoint from the training scenes and packaged in the same format as their training counterparts: - **Puffin-Cam-15M-Bench** β€” **600** single-view image–caption–camera pairs (`.jpg` ↔ `.json`, one shard each). Captions keep the scene-and-camera description; instead of `camera_height`, each JSON records the image `width` / `height`. - **Puffin-Traj-1M-Bench** β€” **100** camera trajectories (90 frames + `cameras.json` per scene, one shard) with the same per-frame annotation schema as Puffin-Traj-1M.

Puffin-16M

| | | |---|---| | **Developed by** | Kang Liao, Yihang Luo, Xiao-Ming Wu, Linyi Jin, Size Wu, Chunyu Lin, Yao Zhao, Fei Wang, Wei Li, Chen Change Loy | | **Affiliations** | S-Lab, Nanyang Technological University; University of Michigan; Beijing Jiaotong University; ACE Robotics | | **First released** | arXiv preprint, 2026 | | **Dataset type** | Camera-centric multimodal dataset for physical-world perception and 3D world modeling | | **Modality** | Image β†’ Text+Camera; Text+Camera β†’ Image; Image+Camera β†’ Image; Image+Camera β†’ Text | --- ## Data Format ### Puffin-Cam-15M β€” captions Each caption is a JSON file, named by a random hash that matches its paired image (`.jpg` ↔ `.json`). It keeps the original scene-and-camera caption and adds a `camera_height` field: ```json { "caption": "The image shows a rocky foreground with a clear blue sky and scattered clouds above ... The camera parameters (roll, pitch, field-of-view, and radial distortion) are: 0.0977, 0.2316, 1.0756, 0.0000.", "camera_height": "Eye-level shot" } ``` `camera_height` is one of: `Eye-level shot`, `Low-position shot`, `High-position shot`, `Aerial shot`, `Underwater shot`. ### Puffin-Traj-1M β€” trajectories Each trajectory scene is a folder (named by a random hash) containing its frames and a single `cameras.json` with per-frame camera annotations: ```json { "motion_type": "pitch", "image_height": 640, "image_width": 640, "vfov_deg": 84.47, "focal_px": 352.49, "images_per_segment": 90, "frames": [ { "file_path": "000001.jpg", "roll_deg": 43.27, "pitch_deg": 44.5, "yaw_deg": 3.38, "camera_intrinsics": [[fx, 0, cx], [0, fy, cy], [0, 0, 1]], "camera_pose": [[ ... 4x4 camera-to-world ... ]] } ] } ``` --- ## Camera Parameters and Ranges ### Puffin-Cam-15M (single-view) Each perspective view is rendered from a panorama with camera parameters sampled as: | Parameter | Range / Distribution | Notes | |---|---|---| | Roll | [βˆ’45Β°, +45Β°] | uniform | | Pitch | [βˆ’45Β°, +45Β°] | uniform | | Vertical FoV | [20Β°, 105Β°] | uniform | | Radial distortion *k₁* | 0 | pinhole camera model (no distortion) | In each caption, the four numbers following *"The camera parameters (roll, pitch, field-of-view, and radial distortion) are:"* are given in **radians**; the radial-distortion term is always 0 under the pinhole model. The paired `camera_height` field further categorizes the shooting height into five levels ordered by increasing altitude, spanning underwater scenes up to aerial and space-like views:

Camera height categories

| `camera_height` | Viewpoint | |---|---| | `Underwater shot` | below the water surface (ocean depths) | | `Low-position shot` | low, near ground / sea level (looking up) | | `Eye-level shot` | human eye level (the most common case) | | `High-position shot` | elevated viewpoint (looking down from height) | | `Aerial shot` | aerial / bird's-eye / space-like overhead view | The altitude figures in the illustration (e.g. *~10 km*, *~50 km*, *100 km+*) are **indicative only** and do not define hard numeric cut-offs for the labels. ### Puffin-Traj-1M (trajectory) Each trajectory is a continuous single-axis camera motion rendered from a panorama: | Parameter | Range / Options | Notes | |---|---|---| | Motion type | `pitch` / `roll` / `yaw` | one axis varies per segment | | Sub-pattern | `mono_pos`, `mono_neg`, `bi_pos_neg`, `bi_neg_pos` | monotonic (up/down, cw/ccw) or bidirectional | | Roll | [βˆ’45Β°, +45Β°] | initial; held constant unless it is the moving axis | | Pitch | [βˆ’45Β°, +45Β°] | initial; held constant unless it is the moving axis | | Yaw | base ∈ [βˆ’180Β°, +180Β°] | swept when `yaw` is the moving axis | | Vertical FoV | [60Β°, 100Β°] | uniform, constant within a trajectory | | Frames / step | 90 frames, 1Β° per frame | each segment covers ~90Β° of motion | Per frame, `cameras.json` stores `roll_deg` / `pitch_deg` / `yaw_deg` (**degrees**), the 3Γ—3 `camera_intrinsics`, and the 4Γ—4 `camera_pose` (camera-to-world). --- ### Directory Structure ``` DATA_PATH/ β”œβ”€ Puffin-Cam-15M/ # 15,338,221 single-view image-caption-camera samples β”‚ β”œβ”€ image_folder/ # perspective images (rendered from panoramas) β”‚ β”‚ β”œβ”€ 000000.zip # 10,000 images per shard, named by random hash β”‚ β”‚ └─ ... # 000000.zip ... 001533.zip β”‚ β”œβ”€ cap_folder/ # captions: scene description + camera parameters + camera height β”‚ β”‚ β”œβ”€ 000000.zip # .json, paired with the same hash in image_folder β”‚ β”‚ └─ ... β”‚ β”œβ”€ summary.json # index mapping each image to its caption β”œβ”€ Puffin-Cam-15M-Bench/ # held-out benchmark: 600 image-caption-camera pairs β”‚ β”œβ”€ image_folder/000000.zip # 600 images, named by hash β”‚ └─ cap_folder/000000.zip # 600 captions, .json paired with image_folder β”œβ”€ Puffin-Traj-1M/ # 1,136,696 camera trajectories (90-frame sequences) β”‚ β”œβ”€ data/ β”‚ β”‚ β”œβ”€ 000000.zip # 500 scenes per shard; each scene: /000001.jpg ... 000090.jpg + cameras.json β”‚ β”‚ └─ ... β”‚ β”œβ”€ summary.json # index of trajectory scenes β”œβ”€ Puffin-Traj-1M-Bench/ # held-out benchmark: 100 trajectories β”‚ └─ data/000001.zip # 100 scenes; each scene: /000001.jpg ... 000090.jpg + cameras.json └─ README.md ``` Both `summary.json` files index the (hashed) samples so that images and their annotations can be located within the shard archives. ### Dataset Download You can download the entire Puffin-16M dataset using the following command: ```bash hf download KangLiao/Puffin-16M --repo-type dataset ``` Puffin-Cam-15M is approximately **2.4 TB**, and Puffin-Traj-1M is approximately **9.2 TB** (~11.6 TB in total); the two held-out benchmarks are small (~100 MB and ~1.1 GB). You may also download a single part, e.g.: ```bash hf download KangLiao/Puffin-16M --repo-type dataset --include "Puffin-Cam-15M/*" # benchmarks only hf download KangLiao/Puffin-16M --repo-type dataset --include "Puffin-Cam-15M-Bench/*" "Puffin-Traj-1M-Bench/*" ``` Pixel-wise camera maps are omitted due to their large total size, but can be generated from the provided camera parameters using ```scripts/camera/cam_dataset.py``` in our [GitHub repository](https://github.com/KangLiao929/Puffin). ### Citation If Puffin-16M or Puffin-World supports your research, please cite our paper: ```bibtex @article{liao2026puffinworld, title = {Puffin-World: Scaling a Unified Multimodal Model with Native 3D World States}, author = {Liao, Kang and Luo, Yihang and Wu, Xiao-Ming and Jin, Linyi and Wu, Size and Lin, Chunyu and Zhao, Yao and Wang, Fei and Li, Wei and Loy, Chen Change}, journal = {arXiv preprint arXiv:2609.04196}, year = {2026} } ``` ### License This project is licensed under [NTU S-Lab License 1.0](https://huggingface.co/datasets/KangLiao/Puffin-16M/blob/main/LICENSE).