MomenTag

About the App

MomenTag is a photo search and management application combining AI-powered semantic search with user tagging features. It helps users organize, search, and discover meaningful moments from their photo collections through intelligent recommendations and natural language processing.

Key Features

Main Features

1. Semantic image search based on user-generated tags

2. Tag recommendations for images

3. Image recommendations for image sets grouped by tags

4. Moment: Tag recommendations for old photos

Detailed Features

Feature	Description
Authentication	JWT-based Sign Up/Log In/Log Out, Token refresh
Photo Upload	Batch processing (8 photos per batch), GPS metadata extraction
Tag Management	Create/Delete user tag albums, Link photos to tags
Semantic Search	Natural language photo search (powered by Qdrant Vector DB)
Hybrid Search	Combined Tag + Semantic search using {TagName} syntax
AI Recommendation	Recommend photos for tags, Recommend tags for photos
Moments	Scrollable photo feed + Dynamic tag recommendations
Auto Captioning	Automatic photo description generation using Vision-Language models

Core Workflow

1. Photo Upload → GPU worker generates image embeddings → Stored in Qdrant
2. Search → Query embedding generated → Vector similarity search → Results returned
3. Recommendation → K-means clustering + Graph analysis → Related photos/tags suggested

Getting Started

Prerequisite

This program uses

• uv for dependency management
• MySQL for database
• Redis for Celery broker
• Qdrant Cloud for vector database

Install above tools properly.

Also, this program targets python >= 3.13. If necessary, install appropriate version with uv.

MySQL Setup

Inside the MySQL prompt on the CPU server, run the following commands. Modify userid and password as you wish.

CREATE DATABASE momentag_db CHARACTER SET utf8mb4 COLLATE utf8mb4_unicode_ci;
CREATE USER 'userid'@'%' IDENTIFIED BY 'password';
GRANT ALL PRIVILEGES ON momentag_db.* TO 'userid'@'%';
FLUSH PRIVILEGES;
EXIT;

Running

Environment Variables

Set the following environment variables on both CPU and GPU servers:

export SECRET_KEY="$YOUR_SECRET_KEY"
export QDRANT_CLIENT_URL="$YOUR_QDRANT_CLOUD_URL"
export QDRANT_API_KEY="$YOUR_QDRANT_API_KEY"
export DJANGO_ALLOWED_HOSTS="$YOUR_IP_ADDRESS"

export DB_HOST="$CPU_SERVER_IP"
export DB_PORT="3306"
export DB_NAME="momentag_db"
export DB_USER="userid"
export DB_PASSWORD="password"

export REDIS_URL="redis://$CPU_SERVER_IP:6379"

CPU Server

cd backend

# Sync dependencies
uv sync

# Run database migrations
uv run manage.py migrate

# Run Django server
uv run manage.py runserver 0.0.0.0:8080

GPU Server

Run the following commands in separate terminals:

cd backend

# Sync dependencies
uv sync

# Run Celery worker for GPU queue (terminal 1)
uv run celery -A config worker -Q gpu -l info --pool=threads -c4

# Run Celery worker for interactive queue (terminal 2)
uv run celery -A config worker -Q interactive -l info --pool=threads -c4

Android

Build and install the android application, then run MomenTag to test features.

Demo Video

https://drive.google.com/file/d/1iEFSSpjw1jEGeSKzshcRyRcf0nCqDglV/view?usp=sharing