AgenticLabeling

AI-Powered Automatic Labeling Platform

Single-service MVP for Florence-2 + SAM2 auto-labeling, review, and export

Features • Architecture • Quick Start • Documentation • License

---

Overview

AgenticLabeling is currently packaged around a single MVP application that handles image upload, Florence-2 detection, SAM2 segmentation, review, and dataset export in one service. The previous microservices stack is preserved only as a legacy deployment path.

Key Capabilities

• Auto-Labeling Pipeline: Image → Florence-2 Detection → SAM2 Segmentation → Registry
• Review Workspace: Browser-based approve/delete flow
• Dataset Export: YOLO and COCO zip export
• Runtime Fallback: If SAM2 is unavailable, the pipeline survives with box-mask fallback
• Legacy Stack Preserved: The old microservices deployment remains available separately

---

Features

AI Models In The MVP

Model	Task	Description
Florence-2	Detection	Open-vocabulary object detection with grounding
SAM2	Segmentation	Instance segmentation with fine masks

Legacy Components Still Available

• DINOv2 classification
• YOLO training
• Video preprocessing / tracking
• Evaluation / MLflow
• Microservice gateway and Streamlit UI

---

Architecture

Default Runtime

Component	Port	GPU	Description
agenticlabeling-mvp	8090	✅	Upload, detect, segment, review, export

Data Flow

Image Input
       │
       ▼
┌──────────────────┐
│    Detection     │ ← Florence-2 grounding
└────────┬─────────┘
         │
         ▼
┌──────────────────┐
│   Segmentation   │ ← SAM2 instance masks
└────────┬─────────┘
         │
         ▼
┌──────────────────┐
│  Local Registry   │ ← SQLite + filesystem assets
└────────┬─────────┘
         │
    ┌────┴────┐
    ▼         ▼
┌───────┐ ┌───────┐
│Review │ │Export │
│Approve│ │YOLO/  │
│Delete │ │COCO   │
└───────┘ └───────┘

Legacy Runtime

The previous multi-container stack is preserved in docker-compose.legacy.yml. The default docker-compose.yml now starts only the MVP app.

---

Quick Start

Prerequisites

• Python 3.10+
• Docker & Docker Compose
• NVIDIA GPU with 8GB+ VRAM (recommended)

Installation

# Start the default MVP container
docker compose up --build

# Or use the helper script
./scripts/run_mvp_docker.sh

# Check health
curl http://localhost:8090/health

Basic Usage

# Auto-label an image
curl -X POST "http://localhost:8090/api/pipeline/auto-label" \
  -F "image=@image.jpg" \
  -F "project_id=default-project" \
  -F "classes=person,car,dog"

# Export dataset
curl -X POST "http://localhost:8090/api/export" \
  -F "dataset_name=my_dataset" \
  -F "export_format=yolo" \
  -F "only_validated=true"

Access UIs

• MVP Home: http://localhost:8090/
• Review UI: http://localhost:8090/review
• API Health: http://localhost:8090/health
• Legacy Stack: ./scripts/run_legacy_docker.sh

---

Documentation

Document	Description
Getting Started	Installation and setup guide
Architecture Spec	Detailed system architecture
Development Progress	Project status and roadmap
PRD	Product requirements document
Navigability Docs	System map, edit map, interface catalog, failure memory
Standards Application	How dev-standards are applied inside this repository
Implementation Memory	Why major implementation choices were made and what they changed
Project Memory	Local engineering memory store and usage

---

API Examples

Auto-Labeling

import httpx

# Label an image
with open("image.jpg", "rb") as f:
    response = httpx.post(
        "http://localhost:8090/api/pipeline/auto-label",
        files={"image": f},
        data={"project_id": "default-project", "classes": "person,car,dog"}
    )
    result = response.json()
    print(f"Detected {result['detections']} objects")

Review Objects

# List objects for a source
response = httpx.get(
    "http://localhost:8090/api/review/objects",
    params={"source_id": "src_1234567890ab"},
)
objects = response.json()["data"]
print(objects[0]["category_name"])

Export Dataset

response = httpx.post(
    "http://localhost:8090/api/export",
    data={
        "dataset_name": "my_dataset",
        "export_format": "yolo",
        "only_validated": "true",
    },
)
export_info = response.json()
print(export_info["download_url"])

---

Testing

# Run the MVP baseline tests
pytest tests/test_mvp_app.py tests/test_mvp_detector.py tests/test_mvp_segmenter.py tests/test_mvp_e2e.py -q

# Run the project-memory regression test
pytest tests/test_project_memory.py -q

---

Project Status

Progress: ████████████████████░ 95%

Phase	Status	Description
Phase 1	✅ Complete	Microservices architecture
Phase 2	✅ Complete	Data management & export
Phase 3	✅ Complete	Video processing & tracking
Phase 4	✅ Complete	Training & evaluation
Phase 5	✅ Complete	Testing (120 tests)
Phase 6	⏳ Pending	Production deployment

---

Contributing

Contributions are welcome! Please read our contributing guidelines before submitting PRs.

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Run tests: pytest tests/ -v
5. Submit a pull request

---

License

This project is licensed under the MIT License - see the LICENSE file for details.

---

Acknowledgments

• Florence-2 - Microsoft
• SAM2 - Meta AI
• DINOv2 - Meta AI
• Ultralytics YOLO - Ultralytics
• X-AnyLabeling - Inspiration for project structure

---

Made with AI-assisted development