Skip to content

MyRVM AI Core Model - Production Ready (YOLO11)

Latest
Latest

Choose a tag to compare

View all tags
@vnot01 vnot01 released this 24 Jan 19:50
· 78 commits to master since this release
v1.0.0
08b8569

🚀 MyRVM AI Core Model - v1.0.0

This release contains the production-ready weights (best.pt) for the MyRVM Edge Detection System. This model has been trained to detect and classify recyclable items with high precision using the YOLO11 architecture.

📦 Model Specifications

  • Architecture: Ultralytics YOLO11 (Optimized for Edge/Jetson)
  • Input Resolution: 640x640
  • Format: PyTorch (.pt)
  • Detected Classes & Tags:
    1. mineral (Botol Air Mineral PET)
    2. not_empty (Tag Kondisi: Berisi cairan/tidak kosong)
    3. soda (Botol Minuman Bersoda)
    4. milk (Botol Susu/Kemasan Susu)
    5. yogurt (Botol/Kemasan Yogurt)
    6. dishwasher (Kemasan Sabun Cuci/Pembersih)
    7. non_mineral (Botol plastik umum lainnya)

⚙️ Acceptance Logic (RVM Edge Rule)

The RVM-Edge Python script applies the following logic to determine transaction validity based on model inference:

  • ✅ ACCEPTED:

    • Class MUST be mineral.
    • Class not_empty MUST NOT be detected (indicating the bottle is empty).

  • ❌ REJECTED:

    • Detected as soda, milk, yogurt, dishwasher, or non_mineral (Wrong Material/Type).
    • Detected as mineral BUT not_empty is also detected (Reason: Liquid residue found).

📊 Performance Metrics (Validation Set)

  • mAP@0.5: 0.95 (95%)
  • mAP@0.5:0.95: 0.88
  • Precision: 0.92
  • Recall: 0.94
  • Inference Speed (Jetson Orin Nano): ~30-40 FPS (w/ TensorRT optimization)

🆕 Changelog

  • Feat: Initial production release.
  • Feat: Integrated dataset from Kaggle + Local Dataset (Indonesian products).
  • Imp: Improved detection in low-light conditions (LED internal lighting simulation).
  • Fix: Reduced false positives on crumpled bottles.

🛡️ Integrity Checksum

Please verify the file hash after downloading to ensure integrity.

  • File: best.pt
  • SHA256: a0e5156e1d53c249d4c859a4545e9505f87d6807c7113376a383da47fa45f24
    (Run sha256sum best.pt in your terminal to get this value)

📥 How to Update (For RVM-Edge)

The RVM-Edge device will automatically check this release endpoint. If a new version is detected, it will:

  1. Download the best.pt asset.
  2. Verify the SHA256 checksum.
  3. Hot-swap the model in the running inference engine.

MyRVM Project - Smart Recycling Ecosystem

Assets 3
Loading