Vestigium-XAI (Phase 0) is an experimental computer vision system designed to classify ammunition calibers based on 2D geometric and morphological features. The application processes optical data to predict the specific caliber type and subsequently maps this prediction to a structured dictionary of compatible firearm platforms.
The core cognitive engine is built upon a Convolutional Neural Network (CNN) architecture. To achieve high accuracy with a lean dataset, the system employs Transfer Learning applied to a ResNet18 backbone, implemented via PyTorch.
Image ingestion, spatial preprocessing, and tensor normalization are handled using OpenCV, ensuring the optical data meets the strict input requirements of the neural network before feature extraction.
The current iteration focuses on establishing the baseline for spatial recognition across four distinct geometric profiles:
- 9x19mm Parabellum
- .45 ACP
- 5.56x45mm NATO
- 12 Gauge Shell
- Python 3.8+
- PyTorch & Torchvision
- OpenCV (
opencv-python) - Pillow
To run an inference test on a local image:
python vestigium.py <path_to_image.jpg>