Skip to content

05-photo-embeddings.org

Latest commit

 

History

History
165 lines (120 loc) · 4.69 KB

05-photo-embeddings.org

File metadata and controls

165 lines (120 loc) · 4.69 KB
Raw

Photo Embeddings and Semantic Search

Introduction

This guide demonstrates using embeddings for semantic photo search.

Setup

First, ensure we have the required model and directories.

# Register the model
llm sentence-transformers register all-MiniLM-L6-v2
# Verify directories
ls -la ../data/images/jwalsh_*  2>/dev/null | wc -l  || echo "Photo directory not found"
ls -la ../data/semantic-photos 2>/dev/null || echo "Semantic photos directory not found"

Photo Semantics Demo

Run the complete photo semantics demo script.

../scripts/photo-semantics.sh

Semantic Search Examples

Test specific semantic searches.

# Food-related searches
llm similar photo-semantics -c "japanese cuisine and sushi"
llm similar photo-semantics -c "pasta dishes with vegetables"
# Urban scenes
llm similar photo-semantics -c "city waterfront or harbor"
# Documentation and UI
llm similar photo-semantics -c "technical documentation or user interface"
# Lifestyle
llm similar photo-semantics -c "coffee shop and reading"

Analyzing Results

Review embedding effectiveness.

# Check embedding collection info
llm collections list | grep photo-semantics

# Get embedding statistics
llm similar photo-semantics --info

# Check individual photo embeddings
for photo in ../data/semantic-photos/*.jpg; do
    echo "Analyzing: $(basename $photo)"
    llm similar photo-semantics -f "$photo" -n 2
    echo "---"
done

Advanced Usage

Example of more complex queries.

Composite queries

llm similar photo-semantics -c "food photography with natural lighting"
llm similar photo-semantics -c "modern urban architecture near water"
llm similar photo-semantics -c "work environment with technology"

NNegative queries (should find few/no matches)

llm similar photo-semantics -c "cars and transportation"
llm similar photo-semantics -c "nature and landscapes"
llm similar photo-semantics -c "portraits and people"

Notes

  • The demo uses the all-MiniLM-L6-v2 model for embeddings
  • Photos are semantically renamed for better organization
  • Results can vary based on the specific photos available
  • Consider experimenting with different embedding models

Interactive Examples

Note: These examples require user interaction and are not suitable for automated execution.

# Interactive photo analysis
llm chat "Analyze this photo" --file ../data/semantic-photos/coffee_and_book.jpg