Back to Projects
Looksense Project

Looksense Machine Learning

Machine Learning

December 2024 - Present

Project Overview

Looksense is a sophisticated machine learning project that implements a 7-class classification model using PyTorch. The model processes input images of size 177x177 pixels and utilizes advanced data augmentation techniques to improve performance. A key technical achievement is the successful conversion of the PyTorch model to TensorFlow.js, enabling browser-based inference. The project is deployed and accessible through both a custom domain (looksense.site) and GitHub Pages.

Technical Implementation

Model Architecture: The PyTorch model is designed to process 177x177 pixel images across 7 distinct classes, providing robust classification capabilities.

Data Augmentation: Implemented advanced augmentation techniques to enhance model generalization and performance, ensuring robust predictions across varied input conditions.

Model Conversion: Successfully converted the PyTorch model to TensorFlow.js through an intermediate TensorFlow conversion, enabling direct browser-based inference without server dependencies.

Technologies Used

PyTorchTensorFlow.jsPythonMachine LearningNext.js Data Augmentation

Key Features

177x177 Input Image Processing
7-Class Classification Model
Advanced Data Augmentation
PyTorch to TensorFlow.js Conversion
Real-time Browser-based Inference
Multiple Deployment Options (looksense.site & GitHub Pages)