AI Scout

AI Scout is a flexible object detection system that combines YOLO with LLMs for intelligent object identification and analysis.

Quick Start

import os
from aiscout import Scout
from aiscout.providers.anthropic import LLM

# Initialize with your preferred LLM
api_key = os.getenv("ANTHROPIC_API_KEY")
llm = LLM(api_key=api_key, model="claude-3-7-sonnet-20250219")
scout = Scout(llm=llm)

# Run detection
result = scout.detect(
    "path/to/image.jpg",
    target_list=["target1", "target2"],
    confidence_threshold=0.2
)

# Save annotated image
result["annotated_image"].save("output.jpg")

Features

• Combines YOLO and LLM capabilities for enhanced object detection
• Supports multiple LLM providers (Anthropic Claude, OpenAI GPT-4V)
• Advanced prompt customization and management
• Iterative refinement with configurable iterations
• Debug mode for development
• Flexible target specification
• Provider-agnostic interface

Requirements

• Python >=3.9
• ultralytics (YOLO)
• requests
• rich
• Anthropic API key (for Claude) or OpenAI API key (for GPT-4V)

Installation

pip install aiscout

Configuration

Set your API key as an environment variable:

# For Anthropic Claude
export ANTHROPIC_API_KEY="your_api_key"
# For OpenAI
export OPENAI_API_KEY="your_api_key"

Advanced Usage

# Enable debug mode
scout = Scout(llm=llm, debug_mode=True)

# Configure detection parameters
result = scout.detect(
    "image.jpg",
    target_list=["target1", "target2"],
    confidence_threshold=0.2,
    min_iterations=3,
    max_iterations=6
)

Prompt Customization

from aiscout.prompts import prompt_manager

# Replace entire prompt
prompt_manager.set_prompt(
    "identify_objects",
    """Analyze this image and identify objects with these requirements:
1. Focus on vehicles and traffic signs
2. Identify make and model when possible
3. Note any safety hazards"""
)

# Append additional instructions
prompt_manager.append_to_prompt(
    "analyze_targets",
    "Additional requirement: Prioritize specific vehicle types over generic classes"
)

# Reset prompts
prompt_manager.reset_prompt("identify_objects")  # Reset specific prompt
prompt_manager.reset_all()  # Reset all prompts

Available prompt types:

• identify_objects: Initial object identification
• analyze_targets: Target analysis and mapping
• refine_detections: Detection refinement rules

Examples

The examples directory contains:

• anthropic/: Claude integration example
• openai/: GPT-4V integration example
• custom_prompts/: Prompt customization examples
• sample_images/: Test images

License

MIT License