MANTRAIDEAS1/nepali-banknote-models

Nepali Banknote Models

A collection of models for the bankNotes-OCR pipeline — a hierarchical 2-stage banknote retrieval system for Nepalese banknotes.

Pipeline Overview

Query Image
  ├── Stage 0: ConvNeXt Binary Classifier  (banknote vs. random)
  ├── Stage 1a: YOLO Denomination Detector  (11 classes)
  ├── Stage 1b: RFDETR + Swin Signature Classifier  (20 governors)
  └── Stage 2: DinoV2-Base patch embeddings + Qdrant vector search

Models in this Repo

1. best_model.pth — ConvNeXt Tiny Binary Classifier

Architecture	convnext_tiny (torchvision)
Purpose	Stage 0 — banknote vs. random image validation
Classes	2: random (0), banknote (1)
Input	224×224 RGB, ImageNet normalization

2. best.pt — YOLO Denomination Detector

Architecture	YOLO (Ultralytics)
Purpose	Stage 1a — detect & classify banknote denomination
Classes	11: 1, 2, 5, 10, 20, 25, 50, 100, 250, 500, 1000
Input	640×640 RGB

3. detector.pth — RFDETR-Large Signature Detector

Architecture	RFDETRLarge (rfdetr)
Purpose	Stage 1b — detect governor signature region on banknote
Detection threshold	0.3

4. classifier.pth — Swin-Base Signature Classifier

Architecture	swin_base_patch4_window7_224 (timm)
Purpose	Stage 1b — classify cropped signature into governor
Classes	20 Nepali governors
Input	224×224 RGB, ImageNet normalization

Governor Classes

Bharat_Raj_Pandey, Bhekh_Bahadur_Thapa, Bijaynath_Bhattarai, Chiranjibi_Nepal, Dipendra_Purush_Dhakal, Ganesh_Bahadur_Thapa, Hari_Shankar_Tripathi, Himalaya_SJB_Rana, Janak_Raj_Pandey, Kalyan_Bikram_Adhikari, Krishna_Bahadur_Manandhar, Kul_Sekhar_Sharma, Laxmi_Nath_Gautam, Maha_Prasad_Adhikari, Narendra_Raj_Pandey, Pradhumna_Lal_Rajbhandari, Satyendra_Pyara_Shrestha, Tilak_Bahadur_Rawal, Yadav_Prasad_Pant, Yubaraj_Khatiwada

Additional Model (not in this repo)

Model	Source	Purpose
DinoV2-Base	facebook/dinov2-base	Stage 2 — patch-level embeddings for visual similarity search

Usage

All models are loaded via huggingface_hub:

from huggingface_hub import hf_hub_download

# ConvNeXt binary classifier
model_path = hf_hub_download(repo_id="MANTRAIDEAS1/nepali-banknote-models", filename="best_model.pth")

# YOLO denomination detector
model_path = hf_hub_download(repo_id="MANTRAIDEAS1/nepali-banknote-models", filename="best.pt")

# RFDETR signature detector
model_path = hf_hub_download(repo_id="MANTRAIDEAS1/nepali-banknote-models", filename="detector.pth")

# Swin signature classifier
model_path = hf_hub_download(repo_id="MANTRAIDEAS1/nepali-banknote-models", filename="classifier.pth")

License

MIT