Update README.md

Comparativa modelos medium, samll and nano

README.md

@@ -16,339 +16,165 @@ pipeline_tag: video-classification
 [![YOLOv8](https://img.shields.io/badge/YOLOv8-Ultralytics-purple.svg)](https://github.com/ultralytics/ultralytics)
 [![Hugging Face](https://img.shields.io/badge/🤗%20Hugging%20Face-Model%20Hub-orange)](https://huggingface.co)
 
-## 📋 Table of Contents
-- [🎯 Quick Comparison](#-quick-comparison)
-- [📊 Model Performance](#-model-performance)  
-- [⚡ Training Efficiency](#-training-efficiency)
-- [🖼️ Visual Results](#️-visual-results)
-- [💰 Resource Usage](#-resource-usage)
-- [🚀 Usage](#-usage)
-- [📈 Detailed Metrics](#-detailed-metrics)
-- [🔗 Downloads](#-downloads)
+# 🚀 Modelos YOLOv8 para Detección de Logos Deportivos - Análisis Comparativo
+### Detección de Objetos de Alto Rendimiento para Reconocimiento de Marcas Deportivas
 
 ---
 
-## 🎯 Quick Comparison
+## 📋 Resumen de Modelos
 
-| Aspect | **YOLOv8 Medium** | **YOLOv8 Small** | **Winner** |
-|--------|-------------------|-------------------|-------------|
-| **🏆 [email protected]** | **97.94%** | **97.04%** | 🥇 Medium (+0.9%) |
-| **📐 Precision** | **97.27%** | **97.22%** | 🥇 Medium (+0.05%) |
-| **🔍 Recall** | **95.74%** | **93.09%** | 🥇 Medium (+2.65%) |
-| **⏱️ Training Time** | 11.55 hours | **4.43 hours** | 🥇 **Small (2.6x faster)** |
-| **💾 Model Size** | ~50 MB | **~25 MB** | 🥇 **Small (50% lighter)** |
-| **🚀 Inference Speed** | ~15ms | **~8-12ms** | 🥇 **Small (25-50% faster)** |
-| **💰 Training Cost** | $57.75 | **$22.15** | 🥇 **Small (62% cheaper)** |
+Este repositorio presenta una comparación exhaustiva de tres modelos YOLOv8 entrenados para detección de logos deportivos, específicamente dirigidos al reconocimiento de **Adidas**, **Nike** y **variantes de Adidas**. Cada modelo ofrece diferentes compensaciones entre precisión, velocidad y requerimientos computacionales.
 
-### 🎖️ **Verdict**: 
-- **🏆 Medium**: Choose for maximum precision (0.9% advantage)
-- **⚡ Small**: Choose for efficiency (2.6x faster training, 50% smaller)
-- **💡 Recommendation**: **Small model** offers the best balance for most applications
+| Modelo | Arquitectura | Parámetros | Tamaño | Velocidad | Mejor Caso de Uso |
+|--------|-------------|------------|---------|-----------|-------------------|
+| **YOLOv8n (Nano)** | Nano | ~3.2M | 6.2 MB | >100 FPS | Móviles/Dispositivos Edge |
+| **YOLOv8s (Small)** | Small | ~11.2M | 21.5 MB | ~80 FPS | Aplicaciones balanceadas |
+| **YOLOv8m (Medium)** | Medium | ~25.9M | 49.7 MB | ~60 FPS | Escenarios alta precisión |
 
 ---
 
-## 📊 Model Performance
+## 🎯 Comparación de Rendimiento
 
-### 🎯 Final Metrics Comparison (Epoch 100)
+### 📊 Resumen de Métricas Clave
 
-| Metric | Medium | Small | Difference | Status |
-|--------|---------|--------|------------|---------|
-| **[email protected]** | 97.94% | 97.04% | -0.90% | 🟡 Minimal loss |
-| **[email protected]:0.95** | 76.43% | 75.19% | -1.24% | 🟡 Minimal loss |
-| **Precision** | 97.27% | 97.22% | -0.05% | 🟢 Negligible |
-| **Recall** | 95.74% | 93.09% | -2.65% | 🟡 Acceptable |
-| **F1-Score** | 96.49% | 95.10% | -1.39% | 🟡 Good trade-off |
-
-### 📈 Performance Evolution
-
-**Training Progress Comparison:**
-```
-Medium Model Evolution:
-Epoch 1:  60.63% → Epoch 100: 97.94% [email protected] (+37.31%)
-
-Small Model Evolution:  
-Epoch 1:  57.58% → Epoch 100: 97.04% [email protected] (+39.46%)
-```
-
-**Key Insights:**
-- ✅ Both models achieve **>97% [email protected]** (excellent performance)
-- ✅ Small model shows **slightly better learning curve** (+39.46% vs +37.31%)
-- ✅ Performance difference is **<1%** - practically negligible
-- ✅ Both models ready for **production deployment**
+| Métrica | Nano (YOLOv8n) | Small (YOLOv8s) | Medium (YOLOv8m) | Ganador |
+|---------|----------------|-----------------|------------------|---------|
+| **[email protected]** | 95.9% | 97.0% | 97.9% | 🥇 Medium |
+| **[email protected]-0.95** | 72.4% | 75.2% | 76.4% | 🥇 Medium |
+| **Precisión** | 98.0% | 97.2% | 97.3% | 🥇 Nano |
+| **Recall** | 92.7% | 93.1% | 95.7% | 🥇 Medium |
+| **F1-Score** | 95.3% | 95.1% | 96.5% | 🥇 Medium |
+| **Tiempo Entrenamiento** | 2.55h | 4.43h | 11.6h | 🥇 Nano |
 
 ---
 
-## ⚡ Training Efficiency
+## 🎨 Recursos Visuales
 
-### ⏱️ Time Comparison
-| Phase | Medium | Small | Improvement |
-|-------|---------|-------|-------------|
-| **Total Training** | 41,586s (11.55h) | 15,966s (4.43h) | **🚀 2.6x faster** |
-| **Per Epoch** | ~415s (6.9min) | ~160s (2.7min) | **⚡ 2.6x faster** |
-| **Convergence** | Epoch 80+ | Epoch 60+ | **📈 20% earlier** |
+### 📊 **Visualizaciones de Rendimiento - Modelo Nano**
 
-### 💾 Resource Requirements
-| Resource | Medium | Small | Advantage |
-|----------|---------|-------|----------|
-| **Model Parameters** | ~25.9M | ~11.2M | Small: 57% less |
-| **VRAM Usage** | ~8GB | ~4-6GB | Small: 25-50% less |
-| **Storage** | ~50MB | ~25MB | Small: 50% smaller |
-| **Inference Memory** | ~200MB | ~100MB | Small: 50% less |
-
----
+#### Evolución de Métricas de Entrenamiento
+![Resultados Nano](./huggingface_assets/nano_results.png)
 
-## 🖼️ Visual Results
+#### Curvas de Rendimiento
+![Curva F1 Nano](./huggingface_assets/nano_boxf1.png)
+![Curva Precisión-Recall Nano](./huggingface_assets/nano_boxpr.png)
 
-### 📊 Training Curves Comparison
+#### Análisis de Confusión - Nano
+![Matriz de Confusión Nano](./huggingface_assets/nano_confusion.png)
 
-| Medium Model | Small Model |
-|:------------:|:-----------:|
-| ![Medium Results](./huggingface_assets/medium_results.png) | ![Small Results](./huggingface_assets/small_results.png) |
-| **97.94% [email protected]** | **97.04% [email protected]** |
+#### Ejemplos de Validación - Nano
+![Etiquetas Validación Nano](./huggingface_assets/nano_val_labels.jpg)
+![Predicciones Validación Nano](./huggingface_assets/nano_val_pred.jpg)
 
-### 🎯 Confusion Matrices
-
-| Medium Model | Small Model |
-|:------------:|:-----------:|
-| ![Medium Confusion](./huggingface_assets/medium_confusion_matrix.png) | ![Small Confusion](./huggingface_assets/small_confusion_matrix.png) |
-| **97.27% Precision** | **97.22% Precision** |
-
-### 📈 Precision-Recall Curves
-
-| Medium Model PR Curve | Small Model PR Curve |
-|:---------------------:|:--------------------:|
-| ![Medium PR](./huggingface_assets/medium_pr_curve.png) | ![Small PR](./huggingface_assets/small_pr_curve.png) |
-
-### 🔍 F1-Score Curves
-
-| Medium F1 Curve | Small F1 Curve |
-|:---------------:|:---------------:|
-| ![Medium F1](./huggingface_assets/medium_f1_curve.png) | ![Small F1](./huggingface_assets/small_f1_curve.png) |
+---
 
-### 📊 Dataset Analysis
+### 📊 **Visualizaciones de Rendimiento - Modelo Small**
 
-| Medium Labels | Small Labels |
-|:-------------:|:------------:|
-| ![Medium Labels](./huggingface_assets/medium_labels.jpg) | ![Small Labels](./huggingface_assets/small_labels.jpg) |
+#### Evolución de Métricas de Entrenamiento
+![Resultados Small](./huggingface_assets/small_results.png)
 
-### 🔬 Validation Examples
+#### Curvas de Rendimiento
+![Curva F1 Small](./huggingface_assets/small_boxf1.png)
+![Curva Precisión-Recall Small](./huggingface_assets/small_boxpr.png)
 
-#### Ground Truth vs Predictions - Batch 0
-| Medium Model | Small Model |
-|:------------:|:-----------:|
-| ![Medium Val GT](./huggingface_assets/medium_val_batch0_labels.jpg) | ![Small Val GT](./huggingface_assets/small_val_batch0_labels.jpg) |
-| ![Medium Val Pred](./huggingface_assets/medium_val_batch0_pred.jpg) | ![Small Val Pred](./huggingface_assets/small_val_batch0_pred.jpg) |
+#### Análisis de Confusión - Small
+![Matriz de Confusión Small](./huggingface_assets/small_confusion.png)
 
-#### Additional Validation Samples
-| Medium Batch 1 | Small Batch 1 |
-|:--------------:|:-------------:|
-| ![Medium Val1](./huggingface_assets/medium_val_batch1_pred.jpg) | ![Small Val1](./huggingface_assets/small_val_batch1_pred.jpg) |
+#### Ejemplos de Validación - Small
+![Etiquetas Validación Small](./huggingface_assets/small_val_labels.jpg)
+![Predicciones Validación Small](./huggingface_assets/small_val_pred.jpg)
 
 ---
 
-## 💰 Resource Usage
+### 📊 **Visualizaciones de Rendimiento - Modelo Medium**
 
-### 💸 Cost Analysis (100 Epochs Training)
+#### Evolución de Métricas de Entrenamiento
+![Resultados Medium](./huggingface_assets/medium_results.png)
 
-| Cost Factor | Medium | Small | Savings |
-|-------------|---------|-------|---------|
-| **GPU Hours** | 11.55h | 4.43h | **-62%** |
-| **Electricity** | ~$4.62 | ~$1.77 | **-62%** |
-| **Cloud Compute** | ~$57.75 | ~$22.15 | **-62%** |
-| **Storage** | $0.50/month | $0.25/month | **-50%** |
-| **Bandwidth** | $1.00 | $0.50 | **-50%** |
+#### Curvas de Rendimiento
+![Curva F1 Medium](./huggingface_assets/medium_boxf1.png)
+![Curva Precisión-Recall Medium](./huggingface_assets/medium_boxpr.png)
 
-### 🌱 Environmental Impact
-- **Carbon Footprint**: Small model = **62% less CO2** emissions
-- **Energy Efficiency**: Small model = **2.6x more efficient**
-- **Sustainability Score**: Small model **⭐⭐⭐⭐⭐** vs Medium **⭐⭐⭐**
+#### Análisis de Confusión - Medium
+![Matriz de Confusión Medium](./huggingface_assets/medium_confusion.png)
 
----
+#### Ejemplos de Validación - Medium
+![Etiquetas Validación Medium](./huggingface_assets/medium_val_labels.jpg)
+![Predicciones Validación Medium](./huggingface_assets/medium_val_pred.jpg)
 
-## 🚀 Usage
+---
 
-### 🔧 Installation
-```bash
-pip install ultralytics torch torchvision
-```
+## 🚀 Ejemplo de Uso
 
-### 📥 Model Download
 ```python
 from ultralytics import YOLO
 
-# Load models
-medium_model = YOLO('path/to/medium/weights/best.pt')
-small_model = YOLO('path/to/small/weights/best.pt')
-```
-
-### 🎯 Inference
-```python
-# Single image prediction
-results_medium = medium_model('image.jpg')
-results_small = small_model('image.jpg')
-
-# Batch prediction
-results_medium = medium_model(['img1.jpg', 'img2.jpg'])
-results_small = small_model(['img1.jpg', 'img2.jpg'])
-```
-
-### 📊 Benchmarking
-```python
-import time
-
-# Speed test
-start = time.time()
-results = small_model('test_image.jpg')
-small_inference_time = time.time() - start
-
-start = time.time()
-results = medium_model('test_image.jpg')
-medium_inference_time = time.time() - start
-
-print(f"Small: {small_inference_time:.3f}s")
-print(f"Medium: {medium_inference_time:.3f}s")
-print(f"Speedup: {medium_inference_time/small_inference_time:.1f}x")
+# Cargar tu modelo preferido
+modelo_nano = YOLO('ruta/a/nano/weights/best.pt')
+modelo_small = YOLO('ruta/a/small/weights/best.pt')  
+modelo_medium = YOLO('ruta/a/medium/weights/best.pt')
+
+# Ejecutar inferencia
+resultados = modelo_nano('ruta/a/tu/imagen.jpg')
+
+# Procesar resultados
+for r in resultados:
+    cajas = r.boxes
+    for caja in cajas:
+        id_clase = caja.cls
+        confianza = caja.conf
+        coordenadas = caja.xyxy
 ```
 
 ---
 
-## 📈 Detailed Metrics
-
-### 📊 Per-Class Performance (Estimated)
-
-| Class | Medium Precision | Small Precision | Medium Recall | Small Recall |
-|-------|------------------|-----------------|---------------|--------------|
-| **Adidas** | ~97.5% | ~97.0% | ~96.0% | ~94.0% |
-| **Adidas_1** | ~97.0% | ~96.8% | ~95.5% | ~92.0% |
-| **Adidas_2** | ~97.2% | ~97.0% | ~95.8% | ~93.0% |
-| **Nike** | ~98.0% | ~97.8% | ~96.0% | ~95.0% |
-
-### 🔄 Convergence Analysis
-
-**Medium Model Convergence:**
-- Epoch 20: 90.68% [email protected]
-- Epoch 60: 96.09% [email protected]  
-- Epoch 100: **97.94% [email protected]**
-
-**Small Model Convergence:**
-- Epoch 20: 88.30% [email protected]
-- Epoch 60: 95.29% [email protected]
-- Epoch 100: **97.04% [email protected]**
-
-### 📉 Loss Evolution
-
-| Loss Type | Medium Final | Small Final | Better |
-|-----------|--------------|-------------|---------|
-| **Box Loss** | 0.7554 | 0.7707 | Medium |
-| **Class Loss** | 0.3743 | 0.4121 | Medium |
-| **DFL Loss** | 1.0088 | 0.9651 | Small |
-
----
-
-## 🔗 Downloads
+## 📊 Información del Dataset
 
-### 📦 Model Weights
-- **Medium Model**: [Download weights](link-to-medium-weights) (50MB)
-- **Small Model**: [Download weights](link-to-small-weights) (25MB)
+### 🎯 Clases Detectadas
+- **adidas**: Variantes principales del logo de Adidas
+- **nike**: Logos swoosh y de texto de Nike
+- **adidas_1**: Diseños alternativos de Adidas
+- **adidas_2**: Variantes especializadas de Adidas
 
-### 📁 Training Artifacts
-- **Medium Configs**: [args.yaml](./runs/detect/modelo_entrenado_medium/args.yaml)
-- **Small Configs**: [args.yaml](./runs/detect/modelo_entrenado_small/args.yaml)
-- **Training Logs**: [results.csv](./runs/detect/)
-
-### 🗂️ Dataset
-- **Sports Logos Dataset**: 4 classes (adidas, adidas_1, adidas_2, nike)
-- **Format**: YOLO format with train/val/test splits
-- **Images**: 416×416px resolution optimized
+### 📸 Estadísticas del Dataset
+- **Total de Imágenes**: 1,200+ muestras
+- **Conjunto de Entrenamiento**: 70% (840+ imágenes)
+- **Conjunto de Validación**: 20% (240+ imágenes)
+- **Conjunto de Prueba**: 10% (120+ imágenes)
+- **Resolución de Imagen**: 416×416 píxeles
+- **Formato de Anotación**: Formato YOLO
 
 ---
 
-## 🎯 Use Case Recommendations
-
-### ✅ Choose **Medium Model** when:
-- **Precision is critical** (forensics, legal applications)
-- **Maximum accuracy needed** (>97.5% required)
-- **Resources are abundant** (cloud/server deployment)
-- **Batch processing** with time flexibility
+## 🎯 Guía de Selección de Modelo
 
-### ⚡ Choose **Small Model** when:
-- **Mobile/edge deployment** required
-- **Real-time performance** needed
-- **Cost optimization** is priority  
-- **Fast training iterations** required
-- **Limited hardware** resources
+### 🚀 Elige **Nano** si necesitas:
+- ✅ Inferencia en tiempo real (>100 FPS)
+- ✅ Despliegue móvil/edge
+- ✅ Consumo mínimo de recursos
+- ✅ Escalado rentable
+- ✅ Dispositivos con batería
 
-### 🏢 Enterprise Recommendations
-- **Production**: Small model for 90% of use cases
-- **Premium tier**: Medium model for high-value clients
-- **Development**: Small model for rapid prototyping
-- **Research**: Medium model for baseline studies
-
----
+### ⚖️ Elige **Small** si necesitas:
+- ✅ Rendimiento/eficiencia balanceados
+- ✅ Flexibilidad de despliegue en cloud
+- ✅ Estabilidad de producción
+- ✅ Requerimientos de precisión moderados
+- ✅ Infraestructura de servidor estándar
 
-## 📊 Technical Specifications
-
-### 🏗️ Architecture Comparison
-| Specification | Medium | Small |
-|---------------|---------|-------|
-| **Base Model** | YOLOv8m | YOLOv8s |
-| **Parameters** | 25.9M | 11.2M |
-| **Layers** | 295 | 225 |
-| **GFLOPs** | 78.9 | 28.6 |
-| **Input Size** | 416×416 | 416×416 |
-
-### ⚙️ Training Configuration
-Both models trained with:
-- **Epochs**: 100
-- **Batch Size**: 4
-- **Optimizer**: AdamW
-- **Learning Rate**: 0.001 → 0.01
-- **Augmentations**: RandAugment, HSV, Mosaic, Erasing
-- **Device**: NVIDIA GPU with CUDA
+### 🎯 Elige **Medium** si necesitas:
+- ✅ Máxima precisión (97.9% [email protected])
+- ✅ Rendimiento grado investigación
+- ✅ Aplicaciones críticas
+- ✅ Capacidades de análisis detallado
+- ✅ Mejor rendimiento de recall (95.7%)
 
 ---
 
-## 📚 Citation
-
-If you use these models in your research, please cite:
-
-```bibtex
-@misc{sports-logo-yolov8-comparison,
-  title={YOLOv8 Sports Logo Detection: Medium vs Small Model Comparison},
-  author={Juan Carlos Macías - Grupo 5},
-  year={2025},
-  publisher={Hugging Face},
-  url={https://huggingface.co/your-username/your-model}
-}
-```
-
----
-
-## 🤝 Contributing
-
-Contributions are welcome! Please feel free to submit a Pull Request.
-
-## 📄 License
-
-This project is licensed under the MIT License - see the [LICENSE](LICENSE) file for details.
-
-## 🙏 Acknowledgments
-
-- [Ultralytics](https://ultralytics.com/) for the YOLOv8 framework
-- Sports brands for logo datasets (educational use)
-- Hugging Face for model hosting platform
-
----
-
-## 📞 Support
-
-- 🐛 [Report Issues](https://github.com/your-repo/issues)
-- 💬 [Discussions](https://github.com/your-repo/discussions)
-- 📧 [Contact](mailto:[email protected])
-
----
+*Última Actualización: 2 de Septiembre, 2025*  
+*Estado de Modelos: ✅ Listos para Producción*  
+*Licencia: MIT*
 
-### 🏆 **Performance Summary: Small Model Delivers 97% of Medium's Accuracy with 2.6x Efficiency Gain**
 
 *Last Updated: September 2, 2025*