---
library_name: transformers
license: apache-2.0
base_model: google/vit-base-patch16-224-in21k
tags:
- image-classification
- vision-transformer
- aquaculture
- fish-disease
- generated_from_trainer
metrics:
- name: accuracy
  type: accuracy
  value: 0.9728
model-index:
- name: fish_disease_datasets
  results:
    - task:
        name: Image Classification
        type: image-classification
      dataset:
        name: fish_disease_datasets
        type: image
      metrics:
        - name: Accuracy
          type: accuracy
          value: 0.9728
---

🐟 Fish Disease Classifier (ViT)

This model is a fine-tuned version of google/vit-base-patch16-224-in21k, trained on a custom fish disease image dataset for Indian aquaculture.
✅ Detected Classes (Fish)

    Bacterial Red Disease

    Bacterial diseases – Aeromoniasis

    Bacterial Gill Disease

    Fungal diseases (Saprolegniasis)

    Parasitic diseases

    Viral diseases (White Tail Disease)

    Healthy Fish

⚠️ Planned Prawn Model (Upcoming)

We are currently working on a separate fine-tuned model to detect:

    Bacterial Gill Disease (BG)

    White Spot Syndrome Virus (WSSV)

    Healthy Prawn

This model will be released in the next version once prawn dataset collection and training is complete.
📊 Evaluation Metrics
Metric	Value
Accuracy	97.28%
Validation Loss	0.0866
Final Epoch	4
🧠 Model Description

    Architecture: Vision Transformer (ViT)

    Base model: google/vit-base-patch16-224-in21k

    Dataset: Custom-labeled images of freshwater fish diseases

    Data augmentation: Albumentations

    Optimized for WhatsApp-based diagnosis tools

🚜 Intended Use

This model is optimized for:

    Farmers needing fast disease detection via image

    WhatsApp or mobile-based advisory tools

    NGO/hatchery/government pilots in India and South Asia

🏋️ Training Summary

    Learning rate: 0.0002

    Batch size: 16 (train) / 8 (eval)

    Epochs: 4

    Mixed Precision: AMP

    Framework: Hugging Face Transformers, PyTorch

### 🏋️ Training Results

| Training Loss | Epoch | Step | Validation Loss | Accuracy |
|---------------|-------|------|-----------------|----------|
| 0.3865        | 0.76  | 100  | 0.4161          | 0.8913   |
| 0.1206        | 1.53  | 200  | 0.2170          | 0.9457   |
| 0.1132        | 2.29  | 300  | 0.1317          | 0.9674   |
| 0.0547        | 3.05  | 400  | 0.0879          | 0.9810   |
| 0.0209        | 3.81  | 500  | 0.0866          | 0.9728   |