xttmp 2.3.0.8

eXtremely Tiny Target - Motion Perception

Small Target Motion Detectors, Version 2.3 (XTT-MP: Extremely Tiny Target - Motion Perception)

See update_log.md for the full version history.

XTT-MP (Extremely Tiny Target - Motion Perception) is a natural architecture-based framework specifically designed for detecting and perceiving the motion of extremely small targets in complex environments.

This release is torch-first: CPU, CUDA, and Apple platforms all use the same torch tensor path, so there is no longer a numpy-only CPU branch to maintain. The model layer now follows a standard nn.Module flow with __init__ and forward, while core blocks keep their internal lifecycle helpers such as setup and reset_buffer.

Built with modularity and extensibility in mind, XTT-MP provides a robust suite of tools for researchers and developers to iterate on tiny-object detection and motion analysis algorithms.

---

✨ Key Features

• Tiny Target Specialist: Optimized feature extraction and attention mechanisms tailored for sub-8x8 pixel objects.
• Motion-Aware Architecture: Integrated spatiotemporal modules to enhance temporal consistency and motion trajectory estimation.
• Decoupled Design: model and core layers are separated cleanly, which makes future learning modules easier to integrate.
• Torch-Only Runtime: inference and post-processing stay in torch tensors on both CPU and GPU, with Apple users able to run the same code path without a CUDA-specific fallback.

📦 Installation

Prerequisites

• Python 3.8+
• PyTorch 1.10+
• CUDA 11.3+ (optional, for NVIDIA acceleration)
• Apple Silicon or macOS users can use the same torch-based workflow without numpy-specific CPU code.

Version Notes

• Model classes now expose the standard __init__ + forward flow.
• Core classes provide lifecycle helpers such as setup and reset_buffer.
• CPU and GPU outputs are both represented as torch tensors.

Example Resources

• Example data and demo assets stay in the GitHub repository under example-data/.
• They are not packaged into the PyPI distribution.
• After pip install xttmp, use the installed code and bring your own input data, or run from a repository checkout to access the bundled examples.

Via PyPI

CPU / MPS

pip install xttmp[torch]

NVIDIA GPU (CUDA 12.6)

pip install xttmp
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu126

Running the GUI Demo

xttmp_gui

Citation

If you find this project useful for your research, please consider citing by this.

@misc{STMDgit,
	author       = {Xu, Mingshuo},
	title        = {Small-Target-Motion-Detectors, Version 2},
	year         = {2024},
	url          = {https://github.com/MingshuoXu/Small-Target-Motion-Detectors},
	note         = {Accessed: 2025-12-12}
}