SmiloAI_model_v2-alpha

85% accuracy trained in yolo

SmiloAI_model_v1-alpha.h5

🧠 SmiloAI_model_v1-alpha.h5

Version

v1-alpha — Experimental / Early Prototype

---

📘 Overview

SmiloAI_model_v1-alpha.h5 is the first experimental model in the SmiloAI series. It represents the initial proof-of-concept stage for the SmiloAI system — a project aimed at combining artificial intelligence with human-centered insight to analyze, interpret, or assist in emotional or behavioral contexts.

As the “alpha” version suggests, this model is in its early development phase, mainly intended for testing, experimentation, and internal validation, not for deployment or production use. Its performance is not yet stable or accurate enough for real-world predictions.

---

⚙️ Technical Details

• Framework: TensorFlow / Keras
• File Type: .h5 (serialized Keras model)
• Architecture: Early prototype neural network (lightweight layers for quick experimentation)
• Stage: Alpha — baseline for structural and accuracy improvements
• Model Purpose: Foundation model for future SmiloAI releases (e.g. v1-beta, v2, etc.)

---

🎯 Intended Use

• Purpose: Experimental research, conceptual testing, and early validation of SmiloAI features
• Users: Developers, students, and researchers analyzing or extending the SmiloAI pipeline
• Not for: Clinical, commercial, or production environments

---

📊 Accuracy and Limitations

This version of the model was trained primarily for experimental validation, not optimized accuracy.
Expected results reflect low to moderate accuracy (typically <70% depending on dataset and task).

Reasons for limited accuracy include:

1. Small or unbalanced dataset during training
2. Minimal hyperparameter tuning
3. Early stopping before convergence
4. Simplified architecture for quick iteration
5. Possible overfitting or underfitting behavior

Thus, SmiloAI_model_v1-alpha.h5 should not be considered reliable for real-world inference.
Its main role is to serve as a structural and functional baseline for improvement.

“Let's Improvise it together... Feel free to discuss some recommendations.”

---

🧪 Evaluation Notes

To check the current performance, run an evaluation script such as:

from tensorflow.keras.models import load_model
from sklearn.metrics import classification_report
import numpy as np

model = load_model('SmiloAI_model_v1-alpha.h5')

# Example placeholders – replace with your real data
X_test = np.load('X_test.npy')
y_test = np.load('y_test.npy')

y_pred_probs = model.predict(X_test)
y_pred = np.argmax(y_pred_probs, axis=1) if y_pred_probs.shape[1] > 1 else (y_pred_probs > 0.5).astype(int).squeeze()

print(classification_report(y_test, y_pred))

**Full Changelog**: https://github.com/nimo94/SmiloAI/compare/SmiloAI-ALPHA...SmiloAI-ALPHA

SmiloAI_model_v1_3a.keras

SmiloAI v1.3

Model File: SmiloAI_model_v1_3a.keras
Release Type: Alpha
Release Date: November 2025

---

Overview

SmiloAI v1.3 delivers a refined and optimized deep learning model for tooth and cavity detection.
Built on the foundation of v1.2, this release enhances diagnostic accuracy, stability, and model efficiency while transitioning fully to the modern .keras format.

---

Key Improvements

• Improved Tooth & Cavity Detection
  Enhanced precision with a reduced false-negative rate and more reliable predictions.

• Upgraded Image Preprocessing
  Cleaner image normalization and scaling pipeline for sharper input analysis.

• Optimized Model Structure
  Redesigned using the latest Keras architecture, offering faster load times and better portability.

• Reduced Model Size
  Efficient layer compression without sacrificing performance or accuracy.

• Faster Inference
  Accelerated processing for both CPU and GPU devices.

---

Technical Notes

• Retrained using an expanded and cleaner dataset with improved labeling consistency.
• Cross-validated for real-world image variation and lighting conditions.

---

📈 Performance Comparison

Metric	v1.2	v1.3
Accuracy	92.7%	95.4%
Inference Time	1.2s	0.8s

---

Future Development (v1.4 Planned)

• Multi-region dental segmentation
• Cavity depth classification
• SmiloAI cloud inference integration

---

Summary

SmiloAI_model_v1_3.keras represents a new level of intelligence, precision, and speed in automated dental imaging analysis.
A smarter tool for researchers, developers, and dental professionals.

---

© 2025 SmiloAI Labs — Sharper Vision, Smarter Diagnosis.

SmiloAI_model_v1_2-alpha.keras

SmiloAI_model_v1_2.keras

Overview

SmiloAI_model_v1_2.keras is the upgraded version of SmiloAI_model_v1-alpha.h5, designed for enhanced facial and smile analysis.
This version introduces improvements in detecting tooth visibility, smile details, and early cavity signs — providing a more accurate and context-aware understanding of facial expressions.

---

Why We Moved from .h5 to .keras

1. Modern Format:
   .keras is now the official Keras model format, offering better reliability and compatibility with TensorFlow 2.11+ and future releases.

2. Complete Model Preservation:
   The new format saves not only weights but also model configuration, training parameters, and custom objects — ensuring consistent loading across environments.

3. Performance Optimization:
   .keras models load and save faster compared to the legacy .h5 format.

4. Future-Proofing:
   The .keras format ensures SmiloAI models remain stable and compatible with upcoming TensorFlow/Keras updates.

---

Accuracy & Detection Improvements

• Enhanced Mouth Region Focus:
  Improved fine-grain detection around the mouth area, making the model more sensitive to tooth visibility and smile structure.

• Cavity Detection:
  Slightly better recognition of potential cavity patterns and tooth irregularities from image input.

• Optimized Preprocessing:
  Better lighting, contrast, and color normalization during training increased robustness to real-world lighting conditions.

Performance Comparison

Feature	v1-alpha (.h5)	v1_2 (.keras)
Format Type	Legacy HDF5	Native Keras
Load Speed	Moderate	Faster
Compatibility	TensorFlow ≤2.10	TensorFlow ≥2.11
Smile/Tooth Detection	Basic	More precise
Cavity Detection	Limited	Improved
Accuracy Gain	—	+3–5% overall

---

Usage

from tensorflow import keras

# Load the model
model = keras.models.load_model("SmiloAI_model_v1_2.keras")

# Make predictions
pred = model.predict(input_image)
print(pred)