Skip to content

end-to-end-detection.md

Latest commit

 

History

History
313 lines (223 loc) · 6.25 KB

end-to-end-detection.md

File metadata and controls

313 lines (223 loc) · 6.25 KB

End-to-End Detection Tutorial

This tutorial walks you through a complete wildlife detection workflow, from images to analysis results — all using the CLI.

Prerequisites

  • WildDetect installed (Installation Guide)
  • Aerial images with wildlife
  • Trained model or access to MLflow registry
  • MLflow server running (optional but recommended)

Workflow Overview

graph LR
    A[Aerial Images] --> B[Configure]
    B --> C[Run Detection]
    C --> D[View Results]
    D --> E[Analyze]
    
    style A fill:#e3f2fd
    style C fill:#fff3e0
    style E fill:#e8f5e9
Loading

Step 1: Prepare Your Environment

1.1 Setup Directory Structure

mkdir D:\wildlife_detection
cd D:\wildlife_detection

# Create directories
mkdir images
mkdir results
mkdir config

1.2 Organize Your Images

D:\wildlife_detection\
├── images\
│   ├── drone_001.jpg
│   ├── drone_002.jpg
│   └── ...
├── results\
└── config\
    └── detection.yaml

1.3 Start MLflow (Optional)

cd wildetect
scripts\launch_mlflow.bat

Access at: http://localhost:5000

Step 2: Configure Detection

2.1 Create Configuration File

Create config/detection.yaml:

model:
  mlflow_model_name: "detector"
  mlflow_model_alias: "production"
  device: "cuda"

image_dir: "D:/wildlife_detection/images/"

processing:
  batch_size: 32
  tile_size: 800
  overlap_ratio: 0.2
  pipeline_type: "simple"  # or "raster" for large images
  confidence_threshold: 0.5

flight_specs:
  flight_height: 120.0
  gsd: 2.38  # Ground Sample Distance (cm/pixel)

output:
  directory: "D:/wildlife_detection/results"
  dataset_name: "my_detections"  # FiftyOne dataset name
  save_visualizations: true

2.2 Alternative: Use Model File

If not using MLflow:

model:
  model_path: "D:/wildlife_detection/models/detector.pt"
  device: "cuda"

2.3 Generate a Default Config

You can generate a starter detection config file using:

wildetect utils create-config detect -o config/detection.yaml

Step 3: Run Detection

Option A: Using the CLI

wildetect detection detect -c config/detection.yaml

Option B: Using the Provided Script

cd wildetect

# Edit config/detection.yaml first
notepad config\detection.yaml

# Run
scripts\run_detection.bat

Expected Output

Processing images: 100%|██████████| 50/50 [00:45<00:00,  1.11it/s]
Detection complete!
Results saved to: D:/wildlife_detection/results/results.json
Total detections: 1,234

Step 4: Review Results

4.1 Results Structure

results/
├── results.json             # All detections with coordinates
└── visualizations/          # Annotated images (if enabled)
    ├── drone_001.jpg
    └── ...

4.2 Detection JSON Format

{
  "image_path": "D:/wildlife_detection/images/drone_001.jpg",
  "image_size": [1920, 1080],
  "processing_time": 0.5,
  "detections": [
    {
      "class_name": "elephant",
      "confidence": 0.95,
      "bbox": [100, 200, 150, 180],
      "bbox_normalized": [0.052, 0.185, 0.078, 0.167]
    }
  ]
}

Step 5: Visualize Results

Option A: Using FiftyOne

# Launch FiftyOne app
wildetect services fiftyone -a launch

# Get dataset info
wildetect services fiftyone -a info -d my_detections

Features:

  • Interactive viewing
  • Filtering by confidence
  • Filtering by species
  • Export capabilities

Option B: View Saved Visualizations

# Open visualizations folder
explorer D:\wildlife_detection\results\visualizations

Option C: Using Web UI

# Launch the Streamlit interface
wildetect services ui

Navigate to the results viewer section.

Step 6: Analyze Results

Run analysis on the detection results:

wildetect detection analyze D:/wildlife_detection/results/results.json \
    -o D:/wildlife_detection/results/analysis/

Step 7: Export to FiftyOne

Export detection results to a FiftyOne dataset for further exploration:

# Export to COCO format
wildetect services fiftyone -a export -d my_detections -f coco -o exports/coco/

# Export to YOLO format
wildetect services fiftyone -a export -d my_detections -f yolo -o exports/yolo/

Step 8: Advanced — Large Raster Detection

For large GeoTIFF / orthomosaic files, use the raster pipeline:

# config/raster_detection.yaml
model:
  mlflow_model_name: "detector"
  device: "cuda"

image_paths:
  - "D:/orthomosaics/large_ortho.tif"

processing:
  tile_size: 800
  overlap_ratio: 0.2
  pipeline_type: "raster"
  nms_threshold: 0.5

flight_specs:
  gsd: 2.38  # Required for raster detection

output:
  directory: "results/raster"
wildetect detection detect -c config/raster_detection.yaml

Step 9: System Info

Check your environment and GPU status:

wildetect utils info

This shows Python version, PyTorch version, CUDA availability, GPU info, and installed dependencies.

Troubleshooting

Detection is Slow

Solutions:

  1. Increase batch size (if GPU memory allows)
  2. Use smaller tile size
  3. Enable GPU acceleration (device: "cuda")
  4. Use the multi-threaded pipeline (pipeline_type: "mt")

Out of Memory

Solutions:

  1. Reduce batch size
  2. Reduce tile size
  3. Use CPU instead of GPU (device: "cpu")
  4. Close other applications

Low Detection Accuracy

Solutions:

  1. Check model is appropriate for your data
  2. Adjust confidence threshold
  3. Verify image quality
  4. Check GSD matches training data

Model Won't Load

Solutions:

  1. Verify MLflow server is running (scripts\launch_mlflow.bat)
  2. Check model name and alias match MLflow registry
  3. Verify model path if using a local file
  4. Check CUDA availability with wildetect utils info

Next Steps

Congratulations! You've completed the end-to-end detection tutorial using the WildDetect CLI. You now know how to configure, run detection, visualize results, and export data.