Project

scene-sorter

Scene Sorter is an AI-powered application that automatically classifies images into real-world scene categories such as mountain, forest, glacier, sea, street, and buildings. It supports both single and batch image uploads, organizes results into class-wise folders, and provides a downloadable ZIP output. Built using MobileNetV2 transfer learning with a FastAPI backend and a Next.js frontend, the project demonstrates a complete, production-ready AI pipeline focused on real-world usability and automation.

Next.js 14TypeScriptTailwind CSSFastAPIPythonUvicorn+8

View Folder Repository Details

Statusshipped

CategoryFrontend: Next.js 14

Stack14 tools

Updated2026-04-06T09:24:11.385368+00:00

Timeline5 steps

Overview

Key Features

Frontend: Next.js 14
TypeScript
Tailwind CSS Backend: FastAPI

Project Timeline

Step 1

Identified the problem of manual image sorting after travel and dataset collection.

Step 2

Built and trained a custom CNN model as a baseline, achieving ~75% accuracy.

Step 3

Migrated to MobileNetV2 transfer learning with fine-tuning, improving accuracy to ~86–87%.

Step 4

Developed a FastAPI backend for single and batch image inference.

Step 5

Built a Next.js frontend with batch upload, folder organization, and ZIP export.

Related Blogs

scene sorter

Scene Sorter is an AI-powered computer vision project that reflects my transition from learning theoretical AI concepts to building practical, real-world systems. The project focuses on solving the problem of manually organizing large image collections by using deep learning to automatically classify images based on their scene type. More than just training a model, Scene Sorter is designed as an end-to-end system — covering data preparation, preprocessing, model design, and inference. It emphasizes thoughtful system design, real-world constraints, and the importance of implementation over is

More Projects

Shipped

Next.js 14TypeScriptTailwind CSSFastAPI+10

scene-sorter

Details →Repo ↗