Update README.md

README.md

@@ -61,39 +61,6 @@ The model serves as a preliminary screening tool to help patients understand pot
 - **Space Demo:** https://huggingface.co/spaces/datdevsteve/nivra-dinov2-medical
 - **Parent Model:** https://huggingface.co/facebook/dinov2-base
 
-## Uses
-
-### Direct Use
-
-This model can be directly used for:
-
-- **Medical Image Screening**: Preliminary classification of skin conditions and visible medical issues
-- **Healthcare Triage Apps**: Assist patients in determining severity and urgency of visible conditions
-- **Telemedicine Platforms**: Support remote healthcare consultations with automated image analysis
-- **Health Education**: Help patients learn about common dermatological conditions
-- **Medical Research**: Analyze patterns in dermatological conditions in Indian populations
-
-### Downstream Use
-
-The model can be integrated into:
-
-- **Mobile Healthcare Applications**: Patient-facing symptom checker apps
-- **Healthcare Chatbots**: Combined with text-based symptom analysis for comprehensive assessment
-- **Clinical Decision Support**: Assist healthcare providers with preliminary image analysis
-- **Public Health Monitoring**: Track prevalence of visible medical conditions
-- **Medical Education Tools**: Training platforms for medical students
-
-### Out-of-Scope Use
-
-❌ **This model should NOT be used for:**
-
-- **Medical Diagnosis**: This is NOT a diagnostic tool and cannot replace professional medical evaluation
-- **Emergency Medical Decisions**: Not suitable for acute or life-threatening situations
-- **Prescription or Treatment Recommendations**: Cannot prescribe medications or treatments
-- **Legal or Insurance Decisions**: Not validated for medicolegal purposes
-- **Surveillance or Non-Medical Purposes**: Not intended for identifying individuals or non-medical use
-- **Conditions Requiring Laboratory Tests**: Cannot detect internal conditions or diseases requiring lab analysis
-
 ## Key Performance Metrics
 
 ### Test Set Results
@@ -147,163 +114,7 @@ Users should:
 - Include emergency contact information for critical cases
 - Log all predictions for quality monitoring and improvement
 
-## How to Get Started with the Model
-
-### Using Transformers Library
-
-```python
-from transformers import AutoImageProcessor, AutoModelForImageClassification
-from PIL import Image
-import torch
-
-# Load model and processor
-model_name = "datdevsteve/nivra-dinov2-finetuned"
-processor = AutoImageProcessor.from_pretrained(model_name)
-model = AutoModelForImageClassification.from_pretrained(model_name)
-
-# Load and process image
-image = Image.open("medical_image.jpg")
-inputs = processor(images=image, return_tensors="pt")
-
-# Get prediction
-with torch.no_grad():
-    outputs = model(**inputs)
-    logits = outputs.logits
-    probabilities = torch.softmax(logits, dim=-1)
-    predicted_class = probabilities.argmax(-1).item()
-    confidence = probabilities[predicted_class].item()
-
-# Get label
-label = model.config.id2label[predicted_class]
-print(f"Condition: {label}")
-print(f"Confidence: {confidence:.2%}")
-```
-
-### Using Pipeline
-
-```python
-from transformers import pipeline
-from PIL import Image
-
-classifier = pipeline("image-classification", model="datdevsteve/nivra-dinov2-finetuned")
-
-image = Image.open("medical_image.jpg")
-results = classifier(image, top_k=3)
-
-for result in results:
-    print(f"{result['label']}: {result['score']:.2%}")
-```
-
-### Via Gradio Space API
-
-```python
-import requests
-import base64
-
-# Read and encode image
-with open("medical_image.jpg", "rb") as f:
-    image_b64 = base64.b64encode(f.read()).decode()
-
-# Call API
-url = "https://datdevsteve-nivra-dinov2-medical.hf.space/api/predict"
-response = requests.post(url, json={"data": [image_b64]})
-
-print(response.json())
-```
-
-## Training Details
-
-### Training Data
-
-- **Dataset Size**: ~12,000 labeled medical images
-- **Image Sources**: 
-  - Curated Indian dermatology databases
-  - Public medical image repositories (DermNet, ISIC Archive - with proper licensing)
-  - Synthetic data augmentation
-- **Conditions Covered**: 40+ common dermatological and visible medical conditions
-- **Image Specifications**:
-  - Resolution: 224x224 pixels (resized)
-  - Format: RGB images
-  - Diverse lighting and quality conditions
-- **Data Split**: 75% train, 15% validation, 10% test
-- **Demographics**: Balanced across different skin tones common in Indian population
-
-### Training Procedure
-
-**Preprocessing:**
-- Image resizing and normalization
-- Data augmentation: rotation, flip, color jitter, brightness adjustment
-- Removal of personally identifiable features (faces blurred when not relevant)
-- Quality filtering (removing blurry or unusable images)
-
-**Training Hyperparameters:**
-
-```python
-{
-  "learning_rate": 5e-5,
-  "batch_size": 32,
-  "num_epochs": 20,
-  "warmup_steps": 500,
-  "weight_decay": 0.01,
-  "optimizer": "AdamW",
-  "lr_scheduler": "cosine",
-  "image_size": 224,
-  "gradient_accumulation_steps": 2
-}
-```
-
-**Training Regime:** Mixed precision (fp16)
-
-**Hardware:**
-- Training Time: ~8 hours
-- GPU: NVIDIA A100 40GB
-- Cloud Provider: Google Cloud Platform
-- Compute Region: asia-south1 (Mumbai)
-
-**Fine-tuning Strategy:**
-- Froze first 6 transformer blocks
-- Fine-tuned last 6 blocks + classification head
-- Classification head: Linear layer (768 → num_classes)
-- Loss function: Cross-entropy with label smoothing (0.1)
-- Early stopping: Patience of 5 epochs
-
-### Evaluation
-
-**Testing Data:**
-- 1,200 held-out medical images
-- Balanced across condition categories
-- Includes challenging cases (multiple conditions, poor lighting)
-
-**Metrics:**
-- Accuracy: Overall correctness
-- Precision/Recall/F1: Per-class and macro-averaged
-- Confusion Matrix: Analyze misclassification patterns
-- Top-3 Accuracy: 96.4% (correct label in top 3 predictions)
-
-## Label Classes
-
-The model classifies medical images into the following categories:
-
-```python
-LABELS = [
-    "Fungal Infection (Ringworm/Tinea)",
-    "Bacterial Skin Infection (Impetigo/Cellulitis)",
-    "Allergic Rash/Contact Dermatitis",
-    "Eczema/Atopic Dermatitis",
-    "Psoriasis",
-    "Acne Vulgaris",
-    "Viral Skin Infection (Warts/Herpes)",
-    "Insect Bite/Sting",
-    "Burn Injury (1st/2nd Degree)",
-    "Wound/Cut/Abrasion",
-    "Urticaria (Hives)",
-    "Seborrheic Dermatitis",
-    "Rosacea",
-    "Vitiligo",
-    "Melanoma (Requires Immediate Medical Attention)",
-    # ... (40+ conditions total)
-]
-```
+
 
 ## Model Architecture and Objective
 
@@ -334,27 +145,10 @@ Carbon emissions estimated using the [Machine Learning Impact calculator](https:
 
 If you use this model in your research or application, please cite:
 
-**BibTeX:**
-
-```bibtex
-@misc{dinov2-indian-medical-2026,
-  author = {Nivra Healthcare Team},
-  title = {DinoV2 for Indian Healthcare Medical Image Classification},
-  year = {2026},
-  publisher = {HuggingFace},
-  journal = {HuggingFace Model Hub},
-  howpublished = {\url{https://huggingface.co/datdevsteve/nivra-dinov2-finetuned}}
-}
-```
-
-**APA:**
-
-Nivra Healthcare Team. (2026). *DinoV2 for Indian Healthcare Medical Image Classification*. HuggingFace Model Hub. https://huggingface.co/datdevsteve/nivra-dinov2-finetuned
 
 ## Model Card Authors
 
 - **datdevsteve** - Model development and fine-tuning
-- **Nivra Team** - Dataset curation, validation, and medical expertise consultation
 
 ## Model Card Contact
 
@@ -368,11 +162,6 @@ For questions, issues, or collaboration inquiries:
 
 ### Version 1.0.0 (January 2026)
 - Initial release
-- 91.2% accuracy on test set
-- Trained on 12,000+ medical images
-- 40+ medical condition classes
-- Optimized for Indian healthcare context
-
 ---
 
 **Last Updated:** January 6, 2026