Anson-Saju-George/wms_dataset

WMS Dataset

Weather Motion Sequences for Deformable Atmospheric Dynamics

Overview

WMS (Weather Motion Sequences) is a large-scale, cleaned dataset designed for video frame interpolation, temporal super-resolution, and motion modeling in atmospheric and satellite imagery.

The dataset focuses on deformable, non-rigid motion patterns such as cloud evolution, dispersion, and atmospheric flow—scenarios where standard natural-video datasets fail to generalize.

This dataset was curated to address domain shift in models trained on human-centric or rigid-motion video benchmarks.

---

Key Characteristics

• Domain: Atmospheric & satellite imagery

• Motion Type: Highly deformable, non-rigid motion (clouds, weather systems)

• Use Cases:

  • Video Frame Interpolation (VFI)
  • Temporal Super-Resolution
  • Spatiotemporal Representation Learning
  • Domain Adaptation & Transfer Learning

• Resolution: Variable (supports arbitrary-resolution training)

• Data Quality: Cleaned, filtered, and sequence-consistent

---

Dataset Statistics

• Total frames: ~9,000
• Dataset downloads: ~1,500+
• Last updated: October 10
• Format: Image sequences suitable for triplet or multi-frame sampling

Note: The statistics reflect active research usage and community adoption.

---

Motivation

Most popular VFI datasets (e.g., Vimeo, UCF, GoPro) are biased toward:

• Rigid or semi-rigid motion
• Human-scale dynamics
• Ground-level scenes

WMS fills the gap by providing:

• Large-scale deformable motion
• Real-world meteorological dynamics
• A challenging domain for generalization and robustness evaluation

---

Recommended Applications

• Domain-adaptive VFI models
• Curriculum learning for motion complexity
• Robust interpolation under large displacement
• Cross-domain benchmarking

---

Citation

If you use this dataset in academic or industrial research, please cite the associated paper:

@article{wms_dataset,
  title   = {Domain-Adapted Video Frame Interpolation for Atmospheric Satellite Imagery},
  author  = {Anson Saju George et al.},
  conference = {IC2NC 2025},
  year    = {2025}
}

---

Links

• Dataset (Hugging Face): https://huggingface.co/datasets/Anson-Saju-George/wms_dataset
• Paper: paper link
• GitHub Repository: https://github.com/Anson-Saju-George/wms-rifev3
• Project Portfolio: https://ansonsajugeorge.online/

---

License

This dataset is released for research and academic use. For commercial usage, please contact the author.

---

Maintainer

Anson Saju George Research focus: Video Frame Interpolation, Domain Adaptation, Deformable Motion Modeling

---

Short Descriptions (Use These Exactly)

📄 Paper Abstract Snippet

We introduce WMS, a domain-specific atmospheric motion dataset designed to evaluate and improve video frame interpolation models under deformable, non-rigid motion. Our experiments demonstrate significant domain shift from natural video benchmarks and highlight the necessity of specialized training data for meteorological imagery.

---

🧠 GitHub Repository Description

Domain-adapted atmospheric motion dataset for video frame interpolation and temporal modeling under deformable dynamics.

---

💼 LinkedIn Project Post

Released WMS, a cleaned atmospheric motion dataset for video frame interpolation and temporal modeling. Built to address domain shift in deformable, non-rigid motion scenarios and validated through domain-adaptive VFI research.

---