Skip to content

LOCAL_DEPLOYMENT.md

Latest commit

 

History

History
346 lines (254 loc) · 8.18 KB

LOCAL_DEPLOYMENT.md

File metadata and controls

346 lines (254 loc) · 8.18 KB

Local Deployment Guide

This guide explains how to deploy the watermark-eraser skill to your local environment, making it ready for AI agents to use directly.

🎯 Goal

After following this guide, you will have:

  • The watermark-eraser skill installed at ~/skills/watermark-eraser/
  • All dependencies (OpenCV, PyTorch, LaMa) installed and ready
  • A working skill that AI agents can call with simple instructions

🚀 Quick Installation (Recommended)

One-Line Install

Run this command to automatically install everything:

/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/moose-lab/watermark-eraser/main/install.sh)"

This will:

  1. Clone the repository to ~/skills/watermark-eraser/
  2. Install all Python dependencies
  3. Verify the installation
  4. Create a test script

Test the Installation

After installation, run:

bash ~/skills/watermark-eraser/test.sh

If you see ✅ Test passed!, the skill is ready to use.

📋 Manual Installation

If you prefer to install manually or need more control:

Step 1: Clone the Repository

# Create the skills directory
mkdir -p ~/skills

# Clone to the correct location
git clone https://github.com/moose-lab/watermark-eraser.git ~/skills/watermark-eraser

Step 2: Install Dependencies

cd ~/skills/watermark-eraser
pip install -r requirements.txt

Dependencies installed:

  • opencv-python - Image processing
  • torch - Deep learning framework
  • simple-lama-inpainting - LaMa inpainting model
  • numpy - Numerical operations

Step 3: Verify Installation

python3 -c "import cv2, torch, simple_lama_inpainting; print('✓ All dependencies installed')"

If this prints ✓ All dependencies installed, you're ready to go.

🤖 Using with AI Agents

Once installed, you can instruct your AI agent to use the skill with simple natural language commands.

Example Instructions

Basic batch processing:

"Use the local watermark-remover skill to process all images in /data/images/ and save them to /data/clean/."

With GPU acceleration:

"Use the watermark-remover skill with GPU acceleration to process the images in /input/ and save to /output/."

With quality assurance:

"Use watermark-remover to process images and save the detection masks for quality checking."

What the Agent Will Do

The agent will automatically run:

python ~/skills/watermark-eraser/src/batch_processor.py \
  --input /data/images/ \
  --output /data/clean/ \
  --batch

And deliver:

  • Processed images (watermark-free)
  • processing_report.json (detailed statistics)
  • output_files.txt (list of output files)

🛠️ Manual Usage

If you want to run the skill manually (without an agent):

Basic Command

python ~/skills/watermark-eraser/src/batch_processor.py \
  --input /path/to/images/ \
  --output /path/to/results/ \
  --batch

Command-Line Options

Option Description Example
--input Input directory or single image /data/images/
--output Output directory or single image /data/clean/
--batch Enable batch processing mode (flag)
--device Processing device (auto/cuda/mps/cpu) --device cuda
--save-masks Save detection masks (flag)
--visualize Save visualization images (flag)

Examples

Process a single image:

python ~/skills/watermark-eraser/src/batch_processor.py \
  --input photo.jpg \
  --output photo_clean.jpg

Batch process with GPU:

python ~/skills/watermark-eraser/src/batch_processor.py \
  --input ./images/ \
  --output ./clean/ \
  --batch \
  --device cuda

Quality assurance workflow:

python ~/skills/watermark-eraser/src/batch_processor.py \
  --input ./images/ \
  --output ./qa_results/ \
  --batch \
  --save-masks \
  --visualize

📊 Output Format

Directory Structure

After processing, your output directory will contain:

output/
├── image1.png              # Processed images
├── image2.png
├── ...
├── processing_report.json  # Detailed statistics
└── output_files.txt        # List of all output files

processing_report.json

{
  "timestamp": "2026-02-22T12:00:00",
  "input_dir": "/data/images/",
  "output_dir": "/data/clean/",
  "total_images": 28,
  "successful": 28,
  "failed": 0,
  "detector_method": "V3 Complete Detector",
  "average_coverage": 3.21,
  "average_time": 5.8,
  "results": [
    {
      "input": "/data/images/1.png",
      "output": "/data/clean/1.png",
      "success": true,
      "mask_coverage": 3.21,
      "processing_time": 5.8
    }
  ]
}

output_files.txt

/data/clean/1.png
/data/clean/2.png
/data/clean/3.png
...

This file list can be used for:

  • Uploading to cloud storage
  • Further image processing
  • Creating archives
  • Integration with other tools

🔧 Troubleshooting

Issue: "Module not found" error

Solution:

cd ~/skills/watermark-eraser
pip install -r requirements.txt

Issue: LaMa model download fails

Solution:

# Manually download the model
mkdir -p ~/.cache/lama
cd ~/.cache/lama
wget https://github.com/enesmsahin/simple-lama-inpainting/releases/download/v0.1.0/big-lama.pt

Issue: GPU out of memory

Solution: Use CPU mode instead:

python ~/skills/watermark-eraser/src/batch_processor.py \
  --input /data/images/ \
  --output /data/clean/ \
  --batch \
  --device cpu

Issue: Detection not accurate

Solution:

  1. Save visualization to check detection: 
    python ~/skills/watermark-eraser/src/batch_processor.py \
  --input /data/images/ \
  --output /data/clean/ \
  --batch \
  --visualize
  2. Check the *_detection.png files to see what was detected
  3. If needed, adjust detection parameters in src/complete_watermark_detector.py

🎓 Advanced Configuration

Customizing Detection Parameters

If you need to process different types of watermarks, you can modify the detector:

# Edit the detector configuration
nano ~/skills/watermark-eraser/src/complete_watermark_detector.py

Key parameters:

  • search_height_ratio: Height of search area (default: 0.15 = bottom 15%)
  • search_width_ratio: Width of search area (default: 0.50 = right 50%)
  • bg_color_lower/upper: Color range for watermark detection
  • min_width/max_width: Expected watermark width range
  • min_height/max_height: Expected watermark height range

Integration with Other Tools

Example: Process and upload to S3

import subprocess
import boto3
from pathlib import Path

# Process images
subprocess.run([
    'python', '~/skills/watermark-eraser/src/batch_processor.py',
    '--input', '/data/images/',
    '--output', '/data/clean/',
    '--batch'
])

# Upload to S3
s3 = boto3.client('s3')
output_files = Path('/data/clean/output_files.txt').read_text().splitlines()

for file_path in output_files:
    s3.upload_file(file_path, 'my-bucket', Path(file_path).name)

📚 Additional Resources

✅ Verification Checklist

After installation, verify that:

  • Repository is cloned to ~/skills/watermark-eraser/
  • All dependencies are installed (python3 -c "import cv2, torch, simple_lama_inpainting")
  • Test script passes (bash ~/skills/watermark-eraser/test.sh)
  • AI agent can find and use the skill
  • Batch processing works on sample images

🎉 You're Ready!

Your watermark-eraser skill is now deployed and ready for AI agents to use. Simply instruct your agent to use the skill, and it will handle the rest automatically.

Example instruction:

"Use the local watermark-remover skill to process all images in /my/images/ and save to /my/clean_images/."

The agent will take care of everything from detection to inpainting to delivering the results!