Subkash2206/spectral-aliasing-cnns

A spectral analysis framework for studying aliasing in strided CNNs. Defines the Alias Violation Ratio (AVR) and Shift Instability Score (SIS) to quantify Nyquist violations and their relationship to prediction instability.

20
/ 100
Experimental

This project provides a framework to understand why image classification models might give different answers when an image is shifted slightly. It takes your trained Convolutional Neural Network (CNN) and images as input, and outputs two scores: Alias Violation Ratio (AVR) and Shift Instability Score (SIS). These scores help machine learning researchers and practitioners understand and quantify how much internal aliasing contributes to prediction instability in their models.

Use this if you are a machine learning researcher or practitioner concerned about the robustness of your CNNs to small input shifts and want to diagnose the underlying causes related to spectral aliasing.

Not ideal if you are looking for an off-the-shelf solution for a specific image classification task without delving into the spectral properties of your model.

deep-learning-robustness image-classification model-diagnostics neural-network-analysis computer-vision
No License No Package No Dependents
Maintenance 13 / 25
Adoption 4 / 25
Maturity 3 / 25
Community 0 / 25

How are scores calculated?

Stars

7

Forks

Language

Python

License

Last pushed

Mar 14, 2026

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/ml-frameworks/Subkash2206/spectral-aliasing-cnns"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.