---
license: cc-by-4.0
task_categories:
  - image-classification
tags:
  - malware-detection
  - 3D-CNN
  - Morton-curve
  - cybersecurity
  - PE-executables
pretty_name: MorVis - 3D Volumetric Malware Tensors
size_categories:
  - 10K<n<100K
---

# MorVis: 3D Volumetric Malware Detection Tensors

## Dataset Description

This dataset contains 6-channel 3D volumetric tensors (64×64×64) generated from Windows PE executables using Morton (Z-order) curve mapping. It accompanies the paper *"3D Volumetric Malware Detection Using Morton Curves and Multi-Channel Semantic Features"* (Harish et al., 2026).

## Dataset Summary

- **Malware tensors**: Generated from the VirusShare_00499 dump (~28,000 samples)
- **Benign tensors**: Not included to save storage and download time (~150GB). Users can generate benign tensors using the provided script on their own installed applications.
- **Tensor shape**: `(6, 64, 64, 64)` per sample, saved as `.npy` files
- **Curve order**: 6 (64³ = 262,144 voxels)

## Channels

Each tensor has 6 semantic channels:

| Channel | Name | Description |
|---------|------|-------------|
| 0 | Raw bytes | Normalized byte values (0–1) |
| 1 | Entropy | Local Shannon entropy over a sliding window |
| 2 | Code mask | Binary mask for executable sections (.text, .code) |
| 3 | Import density | Proximity to import/IAT tables (behavioral signal) |
| 4 | String density | Fraction of printable ASCII in a local window |
| 5 | Data mask | Binary mask: 1 = real file bytes, 0 = padding |

## Generation Script

`malware_3d_multichannel.py` is provided in this repository. Usage:
```bash
python malware_3d_multichannel.py -i ./samples -o ./tensors --order 6
```

**Arguments:**
- `--input_dir / -i`: Directory containing PE files
- `--output_dir / -o`: Output directory for `.npy` tensors
- `--order`: Curve order (default: 6, giving 64³ grid)
- `--min_size`: Minimum file size in KB (default: 10)
- `--max_size`: Maximum file size in MB (default: 50)

The script parses PE headers, extracts relevant sections (skipping resources, relocations, debug), computes all 6 channels, maps bytes into 3D via Morton curve, and saves each tensor as a NumPy `.npy` file along with a `metadata.json`.

## Generating Benign Tensors

Benign tensors are not hosted due to the prohibitive size (~150GB). To generate your own, run the script on locally installed applications:
```bash
python malware_3d_multichannel.py -i "C:\Windows\System32" -o ./tensors_benign
python malware_3d_multichannel.py -i "C:\Program Files" -o ./tensors_benign
```

Any directory containing legitimate PE executables will work.

## Source Data

- **Malware**: VirusShare_00499 dump (Windows PE executables)
- **Benign**: User-installed Windows applications and system files

## Citation

If you use this dataset, please cite (paper currently under review):
```bibtex
@article{harish2026morvis,
  title={3D Volumetric Malware Detection Using Morton Curves and Multi-Channel Semantic Features},
  author={Harish, Parikshieth and P.S., Ramesh and C, Suganthan},
  year={2026}
}
```

## Authors

- Parikshieth Harish (parikshieth.harish2023@vitstudent.ac.in)
- Ramesh P.S. — Corresponding Author (ramesh.ps@vit.ac.in)
- Suganthan C (suganthan.c@vit.ac.in)

School of Computer Science and Engineering, Vellore Institute of Technology, Vellore, India