đŦTechnical Deep Dive
Full Specifications [+]
âī¸ Nexus Index V2.0
đŦ Index Insight
FNI V2.0 for M3arsSynth: Semantic (S:0), Authority (A:0), Popularity (P:0), Recency (R:0), Quality (Q:0).
Verification Authority
đī¸ Data Preview
Row-level preview not available for this dataset.
Schema structure is shown in the Field Logic panel when available.
đ Explore Full Dataset âđ§Ŧ Field Logic
Schema not yet indexed for this dataset.
Dataset Specification
Martian World Model: Controllable Video Synthesis with Physically Accurate 3D Reconstructions
M3arsSynth Dataset Summary
M3arsSynth is a large-scale, multimodal dataset designed for controllable Martian video synthesis. Derived from real NASA Planetary Data System (PDS) imagery, the dataset features over 10,000 physically-accurate 3D scenes rendered into video clips. Each clip is paired with its corresponding camera trajectory and rich textual descriptions, enabling the training of sophisticated, controllable generative models.
Dataset Structure and Content
The dataset is organized into training and validation splits, which are defined by a list of samples in the annotation.json file. The associated media files are stored in corresponding directories.
annotation.json: A JSON file containing a list of data samples for bothtrainandvalsplits. This file is the main entry point for the dataset.train_video/&val_video/: Directories containing the rendered.mp4video clips.train_camera/&val_camera/: Directories containing.txtfiles with the corresponding 6-DOF camera poses for each video.
Each entry in annotation.json is a dictionary that represents one data sample. The fields are described below:
| Field Name | Description |
|---|---|
clip_name |
A unique string identifier for the data sample. |
clip_path |
The relative path to the video file. |
pose_file |
The relative path to the camera pose file. |
terrain_category |
A string classifying the Martian terrain type. |
content_description |
A detailed paragraph describing the visual content of the scene. |
Dataset Creation
The dataset was created by applying a state-of-the-art data curation pipeline to stereo navigation images from NASA's PDS. The process involved:
Rigorous automated and semi-automated filtering to select high-quality images.
A metric-aware 3D reconstruction process to create physically-accurate 3D scenes using 3D Gaussian Splatting.
Rendering of video clips and camera trajectories by sampling virtual camera paths within the reconstructed 3D environments.
Annotation with textual descriptions of scene content (generated by a Vision Language Model).
Licensing Information
The source NASA PDS imagery is in the public domain.
The M3arsSynth dataset, including the curated videos, camera poses, and textual descriptions, is licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0) License. This means you are free to use, share, and adapt the dataset for any purpose, provided you give appropriate credit by citing the original paper.
AI Summary: Based on Hugging Face metadata. Not a recommendation.
đĄī¸ Dataset Transparency Report
Verified data manifest for traceability and transparency.
đ Identity & Source
- id
- hf-dataset--1953493957a--m3arssynth
- slug
- 1953493957a--m3arssynth
- source
- huggingface
- author
- 1953493957a
- license
- CC-BY-4.0
- tags
- license:cc-by-4.0, modality:text, modality:video, arxiv:2507.07978, region:us
âī¸ Technical Specs
- architecture
- null
- params billions
- null
- context length
- null
- pipeline tag
đ Engagement & Metrics
- downloads
- 32,074
- stars
- 0
- forks
- 0
Free2AITools Constitutional Data Pipeline: Curated disclosure mode active. (V15.x Standard)