brohrer/sharpened-cosine-similarity
An alternative to convolution in neural networks
This project offers an alternative to traditional convolution operations in neural networks, designed to efficiently extract features from image data. It takes raw image inputs and processes them through a sharpened cosine similarity mechanism, producing rich feature maps. This tool is intended for machine learning engineers and researchers who are building image classification, object detection, or other computer vision models and are seeking ways to reduce model size without a significant drop in accuracy.
261 stars. No commits in the last 6 months.
Use this if you are developing computer vision models for deployment on resource-constrained devices, where minimizing the number of model parameters is critical.
Not ideal if your primary goal is to achieve state-of-the-art accuracy records or if you need the fastest possible training and inference speeds on high-end GPUs.
Stars
261
Forks
37
Language
Python
License
MIT
Category
Last pushed
Mar 28, 2024
Commits (30d)
0
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/ml-frameworks/brohrer/sharpened-cosine-similarity"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.
Higher-rated alternatives
Dobiasd/frugally-deep
A lightweight header-only library for using Keras (TensorFlow) models in C++.
flashlight/flashlight
A C++ standalone library for machine learning
wichtounet/dll
Fast Deep Learning Library (DLL) for C++ (ANNs, CNNs, RBMs, DBNs...)
NVlabs/tiny-cuda-nn
Lightning fast C++/CUDA neural network framework
KasperskyLab/knp
Kaspersky Neuromorphic Platform