Upload 7 files
Browse files- README.md +76 -19
- app.py +404 -0
- generate_cache.py +22 -0
- logic.py +1470 -0
- packages.txt +1 -0
- requirements.txt +14 -2
- vectores_cache.pkl +3 -0
README.md
CHANGED
|
@@ -1,19 +1,76 @@
|
|
| 1 |
-
-
|
| 2 |
-
|
| 3 |
-
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# DesignIA - Recomendador de Muebles Inteligente
|
| 2 |
+
|
| 3 |
+
DesignIA es una aplicación inteligente que te ayuda a diseñar tu salón ideal. A partir de una imagen panorámica (360°) de tu habitación vacía, la aplicación detecta automáticamente la geometría del espacio (paredes, puertas, ventanas) y te sugiere una distribución óptima de muebles, recomendándote productos reales de IKEA que se ajustan a tu estilo y presupuesto.
|
| 4 |
+
|
| 5 |
+
## Características
|
| 6 |
+
|
| 7 |
+
* **Escaneo de Habitación 3D**: Utiliza **HorizonNet** para detectar la estructura de la habitación a partir de una sola imagen panorámica.
|
| 8 |
+
* **Diseño Automático**: Algoritmos de optimización espacial para colocar muebles respetando zonas de paso y distancias de visualización (TV-Sofá).
|
| 9 |
+
* **Recomendación Estilística**: Un modelo basado en **BERT** analiza el estilo de los muebles para sugerir combinaciones coherentes.
|
| 10 |
+
* **Visualización Interactiva**: Visualiza tu futuro salón en 3D directamente en el navegador.
|
| 11 |
+
* **Presupuesto Ajustable**: Define cuánto quieres gastar y la IA buscará la mejor combinación calidad/precio.
|
| 12 |
+
|
| 13 |
+
## Requisitos Previos
|
| 14 |
+
|
| 15 |
+
Debido al uso de modelos de Inteligencia Artificial avanzados, este proyecto requiere descargar archivos de gran tamaño.
|
| 16 |
+
|
| 17 |
+
1. **Python 3.8+**
|
| 18 |
+
2. **Git LFS (Large File Storage)**: Imprescindible para descargar los modelos de IA.
|
| 19 |
+
* Instalación: [https://git-lfs.com/](https://git-lfs.com/)
|
| 20 |
+
* O ejecuta: `git lfs install`
|
| 21 |
+
|
| 22 |
+
## Instalación
|
| 23 |
+
|
| 24 |
+
1. **Clonar el repositorio**
|
| 25 |
+
Asegúrate de tener Git LFS instalado antes de clonar.
|
| 26 |
+
```bash
|
| 27 |
+
git lfs install
|
| 28 |
+
git clone https://github.com/agerhund/DesignIA.git
|
| 29 |
+
cd DesignIA
|
| 30 |
+
```
|
| 31 |
+
*Nota: La descarga puede tardar unos minutos debido a los modelos (~1.7 GB).*
|
| 32 |
+
|
| 33 |
+
2. **Crear un entorno virtual (Recomendado)**
|
| 34 |
+
```bash
|
| 35 |
+
python -m venv venv
|
| 36 |
+
# En Windows:
|
| 37 |
+
venv\Scripts\activate
|
| 38 |
+
# En Mac/Linux:
|
| 39 |
+
source venv/bin/activate
|
| 40 |
+
```
|
| 41 |
+
|
| 42 |
+
3. **Instalar dependencias**
|
| 43 |
+
```bash
|
| 44 |
+
pip install -r requirements.txt
|
| 45 |
+
```
|
| 46 |
+
|
| 47 |
+
## Ejecución
|
| 48 |
+
|
| 49 |
+
Para iniciar la aplicación web localmente:
|
| 50 |
+
|
| 51 |
+
```bash
|
| 52 |
+
streamlit run app.py
|
| 53 |
+
```
|
| 54 |
+
|
| 55 |
+
La aplicación se abrirá automáticamente en tu navegador (usualmente en `http://localhost:8501`).
|
| 56 |
+
|
| 57 |
+
## Notas sobre el Rendimiento
|
| 58 |
+
|
| 59 |
+
* **Memoria RAM**: Se recomienda disponer de al menos **8 GB de RAM**, ya que los modelos de visión y lenguaje se cargan en memoria.
|
| 60 |
+
* **GPU**: Si dispones de una GPU NVIDIA (CUDA) o un Mac con chip M-series (MPS), la aplicación intentará usarla para acelerar la detección. De lo contrario, funcionará en CPU (más lento).
|
| 61 |
+
* **Streamlit Cloud**: Es posible que esta aplicación **no funcione** en la capa gratuita de Streamlit Cloud debido a las limitaciones de memoria y almacenamiento (los modelos exceden el límite habitual). Se recomienda ejecutar en local o en un servidor con mayores recursos.
|
| 62 |
+
|
| 63 |
+
## Estructura del Proyecto
|
| 64 |
+
|
| 65 |
+
* `app.py`: Punto de entrada de la aplicación Streamlit.
|
| 66 |
+
* `logic.py`: Lógica principal (IA, geometría, recomendación).
|
| 67 |
+
* `models/`: Contiene los pesos de los modelos (HorizonNet y BERT Encoder).
|
| 68 |
+
* `data/`: Base de datos de muebles (CSV).
|
| 69 |
+
* `horizonnet/`: Código fuente del modelo de visión computacional.
|
| 70 |
+
|
| 71 |
+
## Autor
|
| 72 |
+
|
| 73 |
+
**Andrés Gerlotti Slusnys**
|
| 74 |
+
Máster de Data Science, Business Analytics y Big Data
|
| 75 |
+
Universidad Complutense de Madrid
|
| 76 |
+
© 2025
|
app.py
ADDED
|
@@ -0,0 +1,404 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import streamlit as st
|
| 2 |
+
import pandas as pd
|
| 3 |
+
import numpy as np
|
| 4 |
+
from PIL import Image
|
| 5 |
+
import os
|
| 6 |
+
|
| 7 |
+
# Importar la lógica principal del proyecto
|
| 8 |
+
import logic
|
| 9 |
+
|
| 10 |
+
# --- CONFIGURACIÓN INICIAL DE LA PÁGINA ---
|
| 11 |
+
st.set_page_config(page_title="DesignIA - Recomendador de Muebles inteligente", layout="wide")
|
| 12 |
+
|
| 13 |
+
# --- CARGAR ESTILOS CSS ---
|
| 14 |
+
def cargar_estilo():
|
| 15 |
+
"""Define y aplica estilos CSS para la UI de Streamlit."""
|
| 16 |
+
st.markdown("""
|
| 17 |
+
<style>
|
| 18 |
+
/* Ocultar elementos de sistema de Streamlit */
|
| 19 |
+
#MainMenu {visibility: hidden;}
|
| 20 |
+
footer {visibility: hidden;}
|
| 21 |
+
header {visibility: hidden;}
|
| 22 |
+
|
| 23 |
+
/* Estilo de Botones (Azul IKEA) */
|
| 24 |
+
div.stButton > button:first-child {
|
| 25 |
+
background-color: #0051ba;
|
| 26 |
+
color: white;
|
| 27 |
+
border-radius: 8px;
|
| 28 |
+
font-weight: bold;
|
| 29 |
+
border: none;
|
| 30 |
+
padding: 0.5rem 1rem;
|
| 31 |
+
}
|
| 32 |
+
div.stButton > button:first-child:hover {
|
| 33 |
+
background-color: #003e8f;
|
| 34 |
+
border: none;
|
| 35 |
+
}
|
| 36 |
+
|
| 37 |
+
/* Estilo de las Tarjetas de Producto */
|
| 38 |
+
.product-card {
|
| 39 |
+
background-color: white;
|
| 40 |
+
padding: 15px;
|
| 41 |
+
border-radius: 10px;
|
| 42 |
+
box-shadow: 0 2px 5px rgba(0,0,0,0.05);
|
| 43 |
+
margin-bottom: 15px;
|
| 44 |
+
border: 1px solid #eee;
|
| 45 |
+
color: black !important; /* Forzar texto negro dentro de la tarjeta blanca */
|
| 46 |
+
}
|
| 47 |
+
</style>
|
| 48 |
+
""", unsafe_allow_html=True)
|
| 49 |
+
|
| 50 |
+
cargar_estilo()
|
| 51 |
+
|
| 52 |
+
# --- CONSTANTES Y RUTAS DE RECURSOS ---
|
| 53 |
+
# Rutas relativas para despliegue
|
| 54 |
+
csv_path = "data/furniture_data.csv"
|
| 55 |
+
model_path = "models/bert_style_encoder.pth"
|
| 56 |
+
horizon_path = "models/horizonnet_model.pth"
|
| 57 |
+
cache_path = "vectores_cache.pkl" # Se generará en el directorio de trabajo
|
| 58 |
+
|
| 59 |
+
# --- SIDEBAR: INFORMACIÓN DEL PROYECTO ---
|
| 60 |
+
with st.sidebar:
|
| 61 |
+
st.title("DesignIA - Asistente de Diseño")
|
| 62 |
+
st.markdown("---")
|
| 63 |
+
st.markdown("**Trabajo de fin de Máster**")
|
| 64 |
+
st.caption("Máster de Data Science, Business Analytics y Big Data")
|
| 65 |
+
st.caption("Universidad Complutense de Madrid")
|
| 66 |
+
st.markdown("---")
|
| 67 |
+
st.markdown("Desarrollado por **Andrés Gerlotti Slusnys**")
|
| 68 |
+
st.markdown("© 2025")
|
| 69 |
+
|
| 70 |
+
# Indicador de estado para debugging
|
| 71 |
+
with st.expander("Estado del Sistema", expanded=False):
|
| 72 |
+
st.success("Motor Gráfico: Activo")
|
| 73 |
+
st.success("Modelo NLP: Cargado")
|
| 74 |
+
if os.path.exists(horizon_path):
|
| 75 |
+
st.success("HorizonNet: Conectado")
|
| 76 |
+
else:
|
| 77 |
+
st.error("HorizonNet: No encontrado")
|
| 78 |
+
|
| 79 |
+
# Sección de Ejemplos
|
| 80 |
+
st.markdown("---")
|
| 81 |
+
st.subheader("📸 Imágenes de Ejemplo")
|
| 82 |
+
st.caption("Descarga estas imágenes para probar la app:")
|
| 83 |
+
|
| 84 |
+
examples_dir = os.path.join(os.path.dirname(__file__), "examples")
|
| 85 |
+
if os.path.exists(examples_dir):
|
| 86 |
+
example_files = [f for f in os.listdir(examples_dir) if f.endswith(('.jpg', '.png'))]
|
| 87 |
+
example_files.sort()
|
| 88 |
+
|
| 89 |
+
selected_example = st.selectbox("Selecciona un ejemplo:", example_files)
|
| 90 |
+
|
| 91 |
+
if selected_example:
|
| 92 |
+
file_path = os.path.join(examples_dir, selected_example)
|
| 93 |
+
with open(file_path, "rb") as file:
|
| 94 |
+
btn = st.download_button(
|
| 95 |
+
label="⬇️ Descargar Imagen",
|
| 96 |
+
data=file,
|
| 97 |
+
file_name=selected_example,
|
| 98 |
+
mime="image/jpeg"
|
| 99 |
+
)
|
| 100 |
+
|
| 101 |
+
# Mostrar miniatura
|
| 102 |
+
st.image(file_path, caption="Vista previa", use_container_width=True)
|
| 103 |
+
else:
|
| 104 |
+
st.info("No hay ejemplos disponibles.")
|
| 105 |
+
|
| 106 |
+
# --- INICIALIZACIÓN DEL ESTADO DE SESIÓN ---
|
| 107 |
+
if 'stage' not in st.session_state: st.session_state.stage = 0
|
| 108 |
+
if 'room_data' not in st.session_state: st.session_state.room_data = None
|
| 109 |
+
if 'muebles_df' not in st.session_state: st.session_state.muebles_df = None
|
| 110 |
+
if 'data_manager' not in st.session_state: st.session_state.data_manager = None
|
| 111 |
+
|
| 112 |
+
# --- 1. CARGA DE DATOS Y MODELOS (Cacheado) ---
|
| 113 |
+
@st.cache_resource
|
| 114 |
+
def init_backend(csv, cache, model):
|
| 115 |
+
"""Inicializa DataManager y carga/genera el DataFrame de muebles."""
|
| 116 |
+
dm = logic.DataManager(csv, cache, model)
|
| 117 |
+
df = dm.cargar_datos()
|
| 118 |
+
return dm, df
|
| 119 |
+
|
| 120 |
+
try:
|
| 121 |
+
if st.session_state.data_manager is None:
|
| 122 |
+
with st.spinner("Cargando base de datos de muebles y modelos IA..."):
|
| 123 |
+
dm, df = init_backend(csv_path, cache_path, model_path)
|
| 124 |
+
st.session_state.data_manager = dm
|
| 125 |
+
st.session_state.muebles_df = df
|
| 126 |
+
st.sidebar.success(f"Base de datos cargada: {len(st.session_state.muebles_df)} items")
|
| 127 |
+
except Exception as e:
|
| 128 |
+
st.error(f"Error cargando datos: {e}")
|
| 129 |
+
st.stop()
|
| 130 |
+
|
| 131 |
+
# --- INTERFAZ PRINCIPAL ---
|
| 132 |
+
st.title("Recomendador de Muebles Inteligente")
|
| 133 |
+
st.markdown("Sube una panorámica, detecta el espacio y obtén el diseño ideal según tu presupuesto.")
|
| 134 |
+
|
| 135 |
+
# --- PASO 1: CARGA DE IMAGEN Y DETECCIÓN ---
|
| 136 |
+
st.header("1. Escaneo de Habitación")
|
| 137 |
+
uploaded_file = st.file_uploader("Sube tu imagen panorámica (360)", type=['jpg', 'png', 'jpeg'])
|
| 138 |
+
|
| 139 |
+
if uploaded_file is not None:
|
| 140 |
+
image = Image.open(uploaded_file)
|
| 141 |
+
st.image(image, caption='Imagen subida', use_container_width=True)
|
| 142 |
+
|
| 143 |
+
if st.button("Analizar la habitación"):
|
| 144 |
+
with st.spinner("Detectando la geometría..."):
|
| 145 |
+
# Guardar temporalmente la imagen para que la librería pueda leerla
|
| 146 |
+
with open("temp_pano.jpg", "wb") as f:
|
| 147 |
+
f.write(uploaded_file.getbuffer())
|
| 148 |
+
|
| 149 |
+
# Instanciar y ejecutar el detector de layout (HorizonNet)
|
| 150 |
+
try:
|
| 151 |
+
detector = logic.RoomLayoutDetector(horizon_path)
|
| 152 |
+
room_data = detector.detect_layout("temp_pano.jpg")
|
| 153 |
+
|
| 154 |
+
# Validación de datos y manejo de fallos
|
| 155 |
+
if room_data is None or not isinstance(room_data, dict) or 'width' not in room_data:
|
| 156 |
+
st.error("**Detección fallida.** El modelo de Computer Vision no pudo extraer las dimensiones ni los obstáculos. Asegúrate de que el modelo HorizonNet está configurado y funcionando correctamente.")
|
| 157 |
+
st.session_state.room_data = None
|
| 158 |
+
st.session_state.stage = 0
|
| 159 |
+
else:
|
| 160 |
+
st.session_state.room_data = room_data
|
| 161 |
+
st.session_state.stage = 1
|
| 162 |
+
st.success("Análisis completado")
|
| 163 |
+
|
| 164 |
+
# Mostrar resultado de la detección de HorizonNet
|
| 165 |
+
st.header("Resultado del análisis visual")
|
| 166 |
+
annotated_image = logic.dibujar_layout_sobre_imagen("temp_pano.jpg", room_data)
|
| 167 |
+
st.image(annotated_image, caption='Análisis de HorizonNet (Vértices, Puertas y Ventanas)', use_container_width=True)
|
| 168 |
+
|
| 169 |
+
except Exception as e:
|
| 170 |
+
st.error(f"Error en detección: {e}. Revisa la configuración del modelo HorizonNet.")
|
| 171 |
+
st.session_state.stage = 0
|
| 172 |
+
|
| 173 |
+
# --- PASO 2: VERIFICACIÓN Y EDICIÓN DE GEOMETRÍA/OBSTÁCULOS ---
|
| 174 |
+
if st.session_state.stage >= 1 and st.session_state.room_data:
|
| 175 |
+
st.header("2. Verificación de Geometría")
|
| 176 |
+
|
| 177 |
+
# Mostrar dimensiones detectadas
|
| 178 |
+
w_m = st.session_state.room_data.get('width', 0.0)
|
| 179 |
+
l_m = st.session_state.room_data.get('length', 0.0)
|
| 180 |
+
|
| 181 |
+
col1, col2 = st.columns(2)
|
| 182 |
+
with col1:
|
| 183 |
+
st.metric("Ancho (m)", f"{w_m:.2f}")
|
| 184 |
+
with col2:
|
| 185 |
+
st.metric("Largo (m)", f"{l_m:.2f}")
|
| 186 |
+
|
| 187 |
+
# Mostrar el diagrama de planta
|
| 188 |
+
st.subheader("Planta de la habitación")
|
| 189 |
+
floor_plan_fig = logic.generar_diagrama_planta(st.session_state.room_data)
|
| 190 |
+
st.pyplot(floor_plan_fig)
|
| 191 |
+
|
| 192 |
+
# Formulario para añadir puertas y ventanas manualmente
|
| 193 |
+
st.subheader("Añadir puertas y ventanas manualmente")
|
| 194 |
+
|
| 195 |
+
polygon_points = st.session_state.room_data.get('polygon_points', [])
|
| 196 |
+
num_walls = len(polygon_points) if polygon_points is not None else 0
|
| 197 |
+
|
| 198 |
+
if num_walls > 0:
|
| 199 |
+
col_a, col_b, col_c, col_d = st.columns(4)
|
| 200 |
+
|
| 201 |
+
with col_a:
|
| 202 |
+
wall_options = [f"Pared {i+1}" for i in range(num_walls)]
|
| 203 |
+
selected_wall = st.selectbox("Seleccionar Pared", wall_options, key="wall_select")
|
| 204 |
+
wall_idx = int(selected_wall.split()[1]) - 1
|
| 205 |
+
|
| 206 |
+
with col_b:
|
| 207 |
+
obstacle_type = st.radio("Tipo", ["Puerta", "Ventana"], key="obs_type")
|
| 208 |
+
|
| 209 |
+
with col_c:
|
| 210 |
+
# Posición normalizada [0.0, 1.0]
|
| 211 |
+
position_pct = st.number_input("Posición (%)", min_value=0.0, max_value=100.0, value=50.0, step=5.0, key="obs_pos")
|
| 212 |
+
|
| 213 |
+
with col_d:
|
| 214 |
+
width_m = st.number_input("Ancho (m)", min_value=0.1, max_value=5.0, value=0.9, step=0.1, key="obs_width")
|
| 215 |
+
|
| 216 |
+
if st.button("Añadir elemento"):
|
| 217 |
+
# Los datos de centro están normalizados
|
| 218 |
+
new_obstacle = {
|
| 219 |
+
'center': [position_pct / 100.0, wall_idx / max(1, num_walls)],
|
| 220 |
+
'width': width_m
|
| 221 |
+
}
|
| 222 |
+
|
| 223 |
+
if obstacle_type == "Puerta":
|
| 224 |
+
st.session_state.room_data['doors'].append(new_obstacle)
|
| 225 |
+
else:
|
| 226 |
+
st.session_state.room_data['windows'].append(new_obstacle)
|
| 227 |
+
|
| 228 |
+
st.success(f"{obstacle_type} añadida a {selected_wall}")
|
| 229 |
+
st.rerun()
|
| 230 |
+
else:
|
| 231 |
+
st.warning("No se detectaron paredes en el polígono.")
|
| 232 |
+
|
| 233 |
+
# Editor de datos para modificar obstáculos detectados/añadidos
|
| 234 |
+
st.subheader("Editar elementos (puertas y ventanas)")
|
| 235 |
+
st.info("Ajusta las coordenadas de los obstáculos. Los valores X/Y están normalizados [0.0, 1.0].")
|
| 236 |
+
|
| 237 |
+
# Preparar datos para st.data_editor
|
| 238 |
+
doors_data = []
|
| 239 |
+
for i, d in enumerate(st.session_state.room_data.get('doors', [])):
|
| 240 |
+
center_y = d['center'][1] if len(d['center']) > 1 else 0
|
| 241 |
+
doors_data.append({"ID": f"P{i}", "Tipo": "Puerta", "Centro X (Norm.)": d['center'][0], "Centro Y (Norm.)": center_y, "Ancho (m)": d['width']})
|
| 242 |
+
|
| 243 |
+
windows_data = []
|
| 244 |
+
for i, w in enumerate(st.session_state.room_data.get('windows', [])):
|
| 245 |
+
center_y = w['center'][1] if len(w['center']) > 1 else 0
|
| 246 |
+
windows_data.append({"ID": f"V{i}", "Tipo": "Ventana", "Centro X (Norm.)": w['center'][0], "Centro Y (Norm.)": center_y, "Ancho (m)": w['width']})
|
| 247 |
+
|
| 248 |
+
all_obstacles = doors_data + windows_data
|
| 249 |
+
df_obs = pd.DataFrame(all_obstacles)
|
| 250 |
+
|
| 251 |
+
col_config = {
|
| 252 |
+
"Centro X (Norm.)": st.column_config.NumberColumn("Centro X (Norm.)", help="Posición horizontal normalizada [0.0, 1.0]", format="%.2f"),
|
| 253 |
+
"Centro Y (Norm.)": st.column_config.NumberColumn("Centro Y (Norm.)", help="Posición vertical normalizada [0.0, 1.0]", format="%.2f"),
|
| 254 |
+
"Ancho (m)": st.column_config.NumberColumn("Ancho (m)", help="Ancho del obstáculo en metros", format="%.2f"),
|
| 255 |
+
}
|
| 256 |
+
|
| 257 |
+
edited_df = st.data_editor(df_obs, num_rows="dynamic", use_container_width=True, column_config=col_config)
|
| 258 |
+
|
| 259 |
+
if st.button("Confirmar geometría"):
|
| 260 |
+
# Reconstruir el diccionario room_data a partir del DataFrame editado
|
| 261 |
+
new_doors = []
|
| 262 |
+
new_windows = []
|
| 263 |
+
for index, row in edited_df.iterrows():
|
| 264 |
+
obj = {'center': [row['Centro X (Norm.)'], row['Centro Y (Norm.)']], 'width': row['Ancho (m)']}
|
| 265 |
+
if row['Tipo'] == 'Puerta': new_doors.append(obj)
|
| 266 |
+
else: new_windows.append(obj)
|
| 267 |
+
|
| 268 |
+
st.session_state.room_data['doors'] = new_doors
|
| 269 |
+
st.session_state.room_data['windows'] = new_windows
|
| 270 |
+
st.session_state.stage = 2
|
| 271 |
+
st.rerun()
|
| 272 |
+
|
| 273 |
+
# --- PASO 3: PRESUPUESTO Y GENERACIÓN DE LAYOUT/RECOMENDACIÓN ---
|
| 274 |
+
if st.session_state.stage >= 2:
|
| 275 |
+
st.header("3. Presupuesto y generación")
|
| 276 |
+
|
| 277 |
+
presupuesto = st.number_input("Presupuesto Máximo (€)", min_value=100.0, value=1000.0, step=100.0)
|
| 278 |
+
|
| 279 |
+
if st.button("Generar diseño"):
|
| 280 |
+
# Convertir dimensiones de m a cm para el LayoutEngine
|
| 281 |
+
w_cm = st.session_state.room_data.get('width', 0.0) * 100
|
| 282 |
+
l_cm = st.session_state.room_data.get('length', 0.0) * 100
|
| 283 |
+
|
| 284 |
+
if w_cm < 200 or l_cm < 200:
|
| 285 |
+
st.error("Las dimensiones de la habitación son demasiado pequeñas (mínimo 2x2m) o no fueron capturadas correctamente.")
|
| 286 |
+
else:
|
| 287 |
+
with st.spinner("Calculando distribución óptima y seleccionando muebles..."):
|
| 288 |
+
# 1. Inicializar motores
|
| 289 |
+
layout_engine = logic.LayoutEngine(st.session_state.data_manager.dimensiones_promedio)
|
| 290 |
+
recommender = logic.Recommender(st.session_state.muebles_df)
|
| 291 |
+
|
| 292 |
+
# 2. Sugerir el pack de muebles base
|
| 293 |
+
pack_sugerido = layout_engine.sugerir_pack(w_cm, l_cm)
|
| 294 |
+
|
| 295 |
+
# 3. Convertir obstáculos a polígonos para el motor
|
| 296 |
+
obs_layout = layout_engine.convertir_obstaculos(
|
| 297 |
+
st.session_state.room_data,
|
| 298 |
+
w_cm, l_cm,
|
| 299 |
+
polygon_points=st.session_state.room_data.get('polygon_points')
|
| 300 |
+
)
|
| 301 |
+
|
| 302 |
+
# 4. Generar el Layout
|
| 303 |
+
layout_plan, constraints, log_msgs = layout_engine.generar_layout(
|
| 304 |
+
w_cm, l_cm,
|
| 305 |
+
pack_sugerido,
|
| 306 |
+
obs_layout,
|
| 307 |
+
polygon_points=st.session_state.room_data.get('polygon_points')
|
| 308 |
+
)
|
| 309 |
+
|
| 310 |
+
# Mostrar Log de Generación
|
| 311 |
+
with st.expander("📝 Detalles de la Generación del Layout", expanded=False):
|
| 312 |
+
for msg in log_msgs:
|
| 313 |
+
if "✅" in msg: st.success(msg)
|
| 314 |
+
elif "❌" in msg: st.error(msg)
|
| 315 |
+
elif "⚠️" in msg: st.warning(msg)
|
| 316 |
+
else: st.text(msg)
|
| 317 |
+
|
| 318 |
+
if not layout_plan:
|
| 319 |
+
st.error("No se pudo generar una distribución válida para este espacio (demasiado pequeño o muchos obstáculos).")
|
| 320 |
+
else:
|
| 321 |
+
# 5. Recomendar productos (Knapsack para optimización de precio/estilo)
|
| 322 |
+
# Las 'constraints' se definen por los muebles que el layout PUDO colocar
|
| 323 |
+
best_combo = recommender.buscar_combinacion(constraints, presupuesto, top_n=1)
|
| 324 |
+
|
| 325 |
+
if not best_combo:
|
| 326 |
+
st.error("No se encontraron muebles que se ajusten al presupuesto y restricciones.")
|
| 327 |
+
else:
|
| 328 |
+
st.session_state.result_layout = layout_plan
|
| 329 |
+
st.session_state.result_items = best_combo[0]['items']
|
| 330 |
+
st.session_state.result_total = best_combo[0]['precio_total']
|
| 331 |
+
st.session_state.result_score = best_combo[0]['score']
|
| 332 |
+
st.session_state.stage = 3
|
| 333 |
+
|
| 334 |
+
# --- PASO 4: RESULTADOS Y VISUALIZACIÓN FINAL ---
|
| 335 |
+
if st.session_state.stage == 3:
|
| 336 |
+
st.divider()
|
| 337 |
+
st.header("Tu salón ideal")
|
| 338 |
+
|
| 339 |
+
# --- VISUALIZACIÓN 3D Interactiva (Plotly) ---
|
| 340 |
+
st.subheader("Visualización 3D Interactiva")
|
| 341 |
+
|
| 342 |
+
# Generar la figura 3D
|
| 343 |
+
fig_plotly = logic.generar_figura_3d_plotly(
|
| 344 |
+
st.session_state.result_layout,
|
| 345 |
+
st.session_state.room_data,
|
| 346 |
+
st.session_state.result_items
|
| 347 |
+
)
|
| 348 |
+
|
| 349 |
+
# Renderizar la figura de Plotly
|
| 350 |
+
st.plotly_chart(fig_plotly, use_container_width=True, theme="streamlit")
|
| 351 |
+
|
| 352 |
+
st.info("💡 Usa el ratón: Clic izquierdo para rotar, rueda para zoom.")
|
| 353 |
+
|
| 354 |
+
st.divider()
|
| 355 |
+
|
| 356 |
+
# --- LISTA DE COMPRA ---
|
| 357 |
+
st.subheader("Lista de Compra")
|
| 358 |
+
|
| 359 |
+
# Totales y Score de Diseño
|
| 360 |
+
c_tot1, c_tot2 = st.columns([2, 1])
|
| 361 |
+
with c_tot1:
|
| 362 |
+
st.markdown("### Total Estimado")
|
| 363 |
+
st.caption(f"Score de Diseño (Estilo + Puntuación Base): {st.session_state.result_score:.2f}/1.0")
|
| 364 |
+
with c_tot2:
|
| 365 |
+
st.markdown(f"### {st.session_state.result_total:.2f}€")
|
| 366 |
+
|
| 367 |
+
st.markdown("---")
|
| 368 |
+
|
| 369 |
+
# Listado de productos
|
| 370 |
+
for item in st.session_state.result_items:
|
| 371 |
+
with st.container():
|
| 372 |
+
c_img, c_info, c_price, c_link = st.columns([1, 2, 1, 1])
|
| 373 |
+
|
| 374 |
+
url = f"https://www.ikea.com/es/es/p/{item.get('Enlace_producto', '')}-{item.get('ID', '')}"
|
| 375 |
+
img_src = item.get('Imagen_principal', '')
|
| 376 |
+
nombre = item['Nombre']
|
| 377 |
+
tipo = item['Tipo_mueble']
|
| 378 |
+
precio = float(item['Precio'])
|
| 379 |
+
|
| 380 |
+
with c_img:
|
| 381 |
+
if img_src:
|
| 382 |
+
st.image(img_src, width=150)
|
| 383 |
+
else:
|
| 384 |
+
st.text("Sin imagen")
|
| 385 |
+
|
| 386 |
+
with c_info:
|
| 387 |
+
st.subheader(nombre)
|
| 388 |
+
st.caption(tipo)
|
| 389 |
+
st.text(item.get('Descripcion', '')[:100] + '...')
|
| 390 |
+
|
| 391 |
+
with c_price:
|
| 392 |
+
st.markdown(f"### {precio:.2f} €")
|
| 393 |
+
|
| 394 |
+
with c_link:
|
| 395 |
+
st.link_button("Ver en IKEA", url)
|
| 396 |
+
|
| 397 |
+
st.divider()
|
| 398 |
+
|
| 399 |
+
if st.button("Reiniciar"):
|
| 400 |
+
for key in ['room_data', 'result_layout', 'result_items', 'result_total', 'result_score']:
|
| 401 |
+
if key in st.session_state:
|
| 402 |
+
del st.session_state[key]
|
| 403 |
+
st.session_state.stage = 0
|
| 404 |
+
st.rerun()
|
generate_cache.py
ADDED
|
@@ -0,0 +1,22 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
import sys
|
| 3 |
+
import logic
|
| 4 |
+
|
| 5 |
+
# Configurar rutas
|
| 6 |
+
base_dir = os.path.dirname(__file__)
|
| 7 |
+
csv_path = os.path.join(base_dir, "data", "furniture_data.csv")
|
| 8 |
+
model_path = os.path.join(base_dir, "models", "bert_style_encoder.pth")
|
| 9 |
+
cache_path = os.path.join(base_dir, "vectores_cache.pkl")
|
| 10 |
+
|
| 11 |
+
print("Iniciando generación de caché...")
|
| 12 |
+
print(f"CSV: {csv_path}")
|
| 13 |
+
print(f"Modelo: {model_path}")
|
| 14 |
+
|
| 15 |
+
# Inicializar DataManager
|
| 16 |
+
dm = logic.DataManager(csv_path, cache_path, model_path)
|
| 17 |
+
|
| 18 |
+
# Esto forzará la generación y guardado del pickle
|
| 19 |
+
df = dm.cargar_datos()
|
| 20 |
+
|
| 21 |
+
print(f"Caché generado exitosamente en: {cache_path}")
|
| 22 |
+
print(f"Total items: {len(df)}")
|
logic.py
ADDED
|
@@ -0,0 +1,1470 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import pandas as pd
|
| 2 |
+
import numpy as np
|
| 3 |
+
import os
|
| 4 |
+
import itertools
|
| 5 |
+
import pickle
|
| 6 |
+
import torch
|
| 7 |
+
import torch.nn as nn
|
| 8 |
+
from transformers import BertModel, BertTokenizer
|
| 9 |
+
from sklearn.metrics.pairwise import cosine_similarity
|
| 10 |
+
import matplotlib.pyplot as plt
|
| 11 |
+
from shapely.geometry import Polygon, LineString, Point
|
| 12 |
+
from tqdm import tqdm
|
| 13 |
+
import sys
|
| 14 |
+
from PIL import Image, ImageDraw
|
| 15 |
+
import plotly.graph_objects as go
|
| 16 |
+
import pywavefront
|
| 17 |
+
|
| 18 |
+
# --- CONFIGURACIÓN DE RUTAS EXTERNAS ---
|
| 19 |
+
# En despliegue, HorizonNet está en un subdirectorio local
|
| 20 |
+
HORIZON_NET_PATH = os.path.join(os.path.dirname(__file__), 'horizonnet')
|
| 21 |
+
MODEL_DIR = os.path.join(os.path.dirname(__file__), 'modelos_3D')
|
| 22 |
+
|
| 23 |
+
if HORIZON_NET_PATH not in sys.path:
|
| 24 |
+
sys.path.append(HORIZON_NET_PATH)
|
| 25 |
+
|
| 26 |
+
try:
|
| 27 |
+
# Importar desde el paquete local horizonnet
|
| 28 |
+
from horizonnet.model import HorizonNet
|
| 29 |
+
from horizonnet.inference import inference
|
| 30 |
+
from horizonnet.misc import utils
|
| 31 |
+
except ImportError as e:
|
| 32 |
+
print(f"Error al importar HorizonNet desde {HORIZON_NET_PATH}")
|
| 33 |
+
print(f"Detalle: {e}")
|
| 34 |
+
# Definición de clase MOCK para evitar un crash de la aplicación
|
| 35 |
+
class RoomLayoutDetector:
|
| 36 |
+
def __init__(self, model_path):
|
| 37 |
+
print("RoomLayoutDetector en modo MOCK (Error de importación)")
|
| 38 |
+
def detect_layout(self, img_path): return None
|
| 39 |
+
|
| 40 |
+
# ==========================================
|
| 41 |
+
# CLASE DE DETECCIÓN DE LAYOUT (HorizonNet)
|
| 42 |
+
# ==========================================
|
| 43 |
+
class RoomLayoutDetector:
|
| 44 |
+
"""
|
| 45 |
+
Clase para la detección de geometría de habitación (floor/ceiling boundaries)
|
| 46 |
+
y corners a partir de una imagen panorámica 360 usando HorizonNet.
|
| 47 |
+
"""
|
| 48 |
+
def __init__(self, model_path):
|
| 49 |
+
# Utiliza MPS (Metal Performance Shaders) si está disponible en M4
|
| 50 |
+
self.device = torch.device('cpu')
|
| 51 |
+
if torch.backends.mps.is_available():
|
| 52 |
+
self.device = torch.device('mps')
|
| 53 |
+
|
| 54 |
+
print(f"Cargando modelo desde: {model_path} en {self.device}")
|
| 55 |
+
try:
|
| 56 |
+
self.net = utils.load_trained_model(HorizonNet, model_path).to(self.device)
|
| 57 |
+
self.net.eval()
|
| 58 |
+
except Exception as e:
|
| 59 |
+
print(f"Error cargando modelo: {e}")
|
| 60 |
+
self.net = None
|
| 61 |
+
|
| 62 |
+
def detect_layout(self, img_path):
|
| 63 |
+
"""Ejecuta la inferencia de HorizonNet y escala los resultados a metros."""
|
| 64 |
+
if self.net is None: return None
|
| 65 |
+
|
| 66 |
+
# 1. Preprocesar imagen
|
| 67 |
+
try:
|
| 68 |
+
img_pil = Image.open(img_path)
|
| 69 |
+
if img_pil.size != (1024, 512):
|
| 70 |
+
img_pil = img_pil.resize((1024, 512), Image.BICUBIC)
|
| 71 |
+
img_ori = np.array(img_pil)[..., :3].transpose([2, 0, 1]).copy()
|
| 72 |
+
x = torch.FloatTensor([img_ori / 255])
|
| 73 |
+
|
| 74 |
+
# 2. Inferencia
|
| 75 |
+
with torch.no_grad():
|
| 76 |
+
cor_id, z0, z1, vis_out = inference(
|
| 77 |
+
net=self.net, x=x, device=self.device,
|
| 78 |
+
flip=False, rotate=[], visualize=False,
|
| 79 |
+
force_cuboid=False, force_raw=False,
|
| 80 |
+
min_v=None, r=0.05
|
| 81 |
+
)
|
| 82 |
+
|
| 83 |
+
# 3. Procesar resultados: Obtener polígono del suelo
|
| 84 |
+
uv = [[float(u), float(v)] for u, v in cor_id]
|
| 85 |
+
floor_points = self._uv_to_floor_polygon(uv, z0, z1)
|
| 86 |
+
|
| 87 |
+
# 4. Calcular dimensiones y escalar a metros
|
| 88 |
+
min_x, min_y = floor_points.min(axis=0)
|
| 89 |
+
max_x, max_y = floor_points.max(axis=0)
|
| 90 |
+
ancho_bbox = max_x - min_x
|
| 91 |
+
largo_bbox = max_y - min_y
|
| 92 |
+
altura_unidades = abs(z1 - z0)
|
| 93 |
+
|
| 94 |
+
# Factor de escala: Se asume altura de cámara de 1.6m (z0)
|
| 95 |
+
ALTURA_CAMARA = 1.6
|
| 96 |
+
factor_escala = ALTURA_CAMARA / z0
|
| 97 |
+
|
| 98 |
+
ancho_m = ancho_bbox * factor_escala
|
| 99 |
+
largo_m = largo_bbox * factor_escala
|
| 100 |
+
altura_m = altura_unidades * factor_escala
|
| 101 |
+
|
| 102 |
+
# Escalar puntos para visualización 3D y normalizar para 2D (planta)
|
| 103 |
+
floor_points_scaled = floor_points * factor_escala
|
| 104 |
+
# Para la planta 2D, normalizamos para que empiece en (0,0) y esté en metros
|
| 105 |
+
floor_points_norm = (floor_points_scaled - [floor_points_scaled[:,0].min(), floor_points_scaled[:,1].min()])
|
| 106 |
+
|
| 107 |
+
# Obtener datos raw para la visualización de la detección
|
| 108 |
+
x_tensor = x.to(self.device)
|
| 109 |
+
y_bon_, y_cor_ = self.net(x_tensor)
|
| 110 |
+
y_bon_ = y_bon_.cpu().detach().numpy()[0]
|
| 111 |
+
y_cor_ = torch.sigmoid(y_cor_).cpu().detach().numpy()[0, 0]
|
| 112 |
+
|
| 113 |
+
return {
|
| 114 |
+
'width': ancho_m,
|
| 115 |
+
'length': largo_m,
|
| 116 |
+
'height': altura_m,
|
| 117 |
+
'doors': [], 'windows': [], # Estos deberían ser inferidos en tu versión completa
|
| 118 |
+
'y_bon': y_bon_, 'y_cor': y_cor_,
|
| 119 |
+
'polygon_points': floor_points_norm.tolist(),
|
| 120 |
+
'polygon_points_raw': floor_points_scaled.tolist(),
|
| 121 |
+
'factor_escala': factor_escala
|
| 122 |
+
}
|
| 123 |
+
|
| 124 |
+
except Exception as e:
|
| 125 |
+
print(f"Error en inferencia: {e}")
|
| 126 |
+
import traceback
|
| 127 |
+
traceback.print_exc()
|
| 128 |
+
return None
|
| 129 |
+
|
| 130 |
+
def _uv_to_floor_polygon(self, uv, z0, z1):
|
| 131 |
+
"""Convierte los puntos uv (coordenadas de la imagen) en coordenadas (x,y) del plano del suelo."""
|
| 132 |
+
floor_points = []
|
| 133 |
+
for i in range(0, len(uv), 2):
|
| 134 |
+
u = uv[i][0]
|
| 135 |
+
v_floor = uv[i+1][1]
|
| 136 |
+
lon = (u - 0.5) * 2 * np.pi
|
| 137 |
+
lat = (0.5 - v_floor) * np.pi
|
| 138 |
+
# Fórmula de proyección esférica
|
| 139 |
+
r = abs(z0 / np.tan(lat)) if np.tan(lat) != 0 else 0
|
| 140 |
+
x = r * np.cos(lon)
|
| 141 |
+
y = r * np.sin(lon)
|
| 142 |
+
floor_points.append([x, y])
|
| 143 |
+
return np.array(floor_points)
|
| 144 |
+
|
| 145 |
+
# ==========================================
|
| 146 |
+
# 1. MODELO DE ESTILO (Encoder basado en BERT)
|
| 147 |
+
# ==========================================
|
| 148 |
+
class StyleEncoder(nn.Module):
|
| 149 |
+
"""
|
| 150 |
+
Encoder de estilo basado en BERT para generar un vector de embedding
|
| 151 |
+
a partir de la descripción de un mueble.
|
| 152 |
+
"""
|
| 153 |
+
def __init__(self, n_dims=128):
|
| 154 |
+
super(StyleEncoder, self).__init__()
|
| 155 |
+
self.bert = BertModel.from_pretrained('bert-base-multilingual-cased')
|
| 156 |
+
self.fc = nn.Linear(self.bert.config.hidden_size, n_dims)
|
| 157 |
+
|
| 158 |
+
def forward(self, input_ids, attention_mask):
|
| 159 |
+
bert_output = self.bert(input_ids=input_ids, attention_mask=attention_mask)
|
| 160 |
+
pooler_output = bert_output[1] # Vector [CLS]
|
| 161 |
+
vector_estilo = self.fc(pooler_output)
|
| 162 |
+
return vector_estilo
|
| 163 |
+
|
| 164 |
+
# ==========================================
|
| 165 |
+
# 2. GESTOR DE DATOS Y CARGA
|
| 166 |
+
# ==========================================
|
| 167 |
+
class DataManager:
|
| 168 |
+
"""Gestiona la carga de la base de datos, la vectorización y el cacheo."""
|
| 169 |
+
def __init__(self, csv_path, vectors_cache_path, bert_model_path):
|
| 170 |
+
self.csv_path = csv_path
|
| 171 |
+
self.cache_path = vectors_cache_path
|
| 172 |
+
self.bert_model_path = bert_model_path
|
| 173 |
+
self.df_muebles = None
|
| 174 |
+
self.dimensiones_promedio = {}
|
| 175 |
+
|
| 176 |
+
# Tipos de muebles relevantes para el layout del salón
|
| 177 |
+
self.muebles_a_usar = [
|
| 178 |
+
'Sofás', 'Sillones', 'Muebles de salón',
|
| 179 |
+
'Mesas bajas de salón, de centro y auxiliares',
|
| 180 |
+
'Estanterías y librerías'
|
| 181 |
+
]
|
| 182 |
+
|
| 183 |
+
def cargar_datos(self):
|
| 184 |
+
"""Carga los vectores desde caché o los genera usando el StyleEncoder."""
|
| 185 |
+
if os.path.exists(self.cache_path):
|
| 186 |
+
print(f"Cargando caché de: {self.cache_path}")
|
| 187 |
+
with open(self.cache_path, 'rb') as f:
|
| 188 |
+
self.df_muebles = pickle.load(f)
|
| 189 |
+
else:
|
| 190 |
+
print("Generando vectores (esto puede tardar)...")
|
| 191 |
+
self._generar_vectores()
|
| 192 |
+
|
| 193 |
+
# Filtrar sofás excesivamente grandes (> 300cm) que rompen el layout en habitaciones normales
|
| 194 |
+
if self.df_muebles is not None:
|
| 195 |
+
self.df_muebles = self.df_muebles[~((self.df_muebles['Tipo_mueble'] == 'Sofás') & (self.df_muebles['Ancho'] > 300))]
|
| 196 |
+
|
| 197 |
+
# Pre-procesar columnas de dimensiones para asegurar que son numéricas y válidas
|
| 198 |
+
if self.df_muebles is not None:
|
| 199 |
+
cols_dim = ['Ancho', 'Largo', 'Altura']
|
| 200 |
+
for col in cols_dim:
|
| 201 |
+
self.df_muebles[col] = pd.to_numeric(self.df_muebles[col], errors='coerce')
|
| 202 |
+
self.df_muebles.loc[self.df_muebles[col] <= 0, col] = np.nan
|
| 203 |
+
|
| 204 |
+
# Calcular dimensiones promedio por categoría
|
| 205 |
+
if self.df_muebles is not None:
|
| 206 |
+
avg_df = self.df_muebles.groupby('Tipo_mueble')[['Ancho', 'Largo']].mean()
|
| 207 |
+
self.dimensiones_promedio = avg_df.to_dict('index')
|
| 208 |
+
|
| 209 |
+
# Normalizar claves a minúsculas para evitar errores de key y aplicar lógica de negocio
|
| 210 |
+
processed_dims = {}
|
| 211 |
+
for k, v in self.dimensiones_promedio.items():
|
| 212 |
+
ancho = round(v.get('Ancho', 100.0), 1)
|
| 213 |
+
largo = round(v.get('Largo', 50.0), 1)
|
| 214 |
+
|
| 215 |
+
# Correcciones de lógica de negocio o datos faltantes (en cm)
|
| 216 |
+
if k == 'Sofás':
|
| 217 |
+
if largo < 50: largo = 90.0
|
| 218 |
+
if k == 'Sillones':
|
| 219 |
+
if largo < 40: largo = ancho if ancho >= 40 else 70.0
|
| 220 |
+
|
| 221 |
+
processed_dims[k] = {'ancho': ancho, 'largo': largo}
|
| 222 |
+
self.dimensiones_promedio = processed_dims
|
| 223 |
+
|
| 224 |
+
# Imputar valores faltantes o inválidos en el DataFrame con los promedios calculados
|
| 225 |
+
if self.df_muebles is not None:
|
| 226 |
+
for tipo, dims in self.dimensiones_promedio.items():
|
| 227 |
+
mask_tipo = self.df_muebles['Tipo_mueble'] == tipo
|
| 228 |
+
|
| 229 |
+
# Imputar Ancho
|
| 230 |
+
mask_invalid_w = mask_tipo & (self.df_muebles['Ancho'].isna())
|
| 231 |
+
if mask_invalid_w.any():
|
| 232 |
+
self.df_muebles.loc[mask_invalid_w, 'Ancho'] = dims['ancho']
|
| 233 |
+
|
| 234 |
+
# Imputar Largo
|
| 235 |
+
mask_invalid_l = mask_tipo & (self.df_muebles['Largo'].isna() | (self.df_muebles['Largo'] <= 0))
|
| 236 |
+
if mask_invalid_l.any():
|
| 237 |
+
self.df_muebles.loc[mask_invalid_l, 'Largo'] = dims['largo']
|
| 238 |
+
|
| 239 |
+
# Imputar Altura (opcional, pero bueno para consistencia)
|
| 240 |
+
# mask_invalid_h = mask_tipo & (self.df_muebles['Altura'].isna() | (self.df_muebles['Altura'] <= 0))
|
| 241 |
+
# if mask_invalid_h.any():
|
| 242 |
+
# self.df_muebles.loc[mask_invalid_h, 'Altura'] = 60 # Valor por defecto seguro
|
| 243 |
+
|
| 244 |
+
return self.df_muebles
|
| 245 |
+
|
| 246 |
+
def _generar_vectores(self, n_dims=128):
|
| 247 |
+
"""Procesa el CSV y genera el vector de estilo para cada mueble."""
|
| 248 |
+
if not os.path.exists(self.csv_path):
|
| 249 |
+
raise FileNotFoundError(f"No se encuentra el CSV en {self.csv_path}")
|
| 250 |
+
|
| 251 |
+
df = pd.read_csv(self.csv_path)
|
| 252 |
+
df_filtrado = df[df['Tipo_mueble'].isin(self.muebles_a_usar)].copy()
|
| 253 |
+
|
| 254 |
+
# Carga modelo BERT
|
| 255 |
+
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
| 256 |
+
model = StyleEncoder(n_dims=n_dims).to(device)
|
| 257 |
+
|
| 258 |
+
if os.path.exists(self.bert_model_path):
|
| 259 |
+
model.load_state_dict(torch.load(self.bert_model_path, map_location=device))
|
| 260 |
+
else:
|
| 261 |
+
print("ADVERTENCIA: No se encontraron pesos del modelo BERT, usando aleatorios.")
|
| 262 |
+
|
| 263 |
+
tokenizer = BertTokenizer.from_pretrained('bert-base-multilingual-cased')
|
| 264 |
+
model.eval()
|
| 265 |
+
|
| 266 |
+
df_filtrado['text_data'] = df_filtrado['Nombre'].fillna('') + ' ' + df_filtrado['Descripcion'].fillna('')
|
| 267 |
+
generated_vectors = []
|
| 268 |
+
|
| 269 |
+
with torch.no_grad():
|
| 270 |
+
for text in tqdm(df_filtrado['text_data'], desc="Vectorizando"):
|
| 271 |
+
tokens = tokenizer(text, padding='max_length', truncation=True, max_length=64, return_tensors='pt')
|
| 272 |
+
input_ids = tokens['input_ids'].to(device)
|
| 273 |
+
attention_mask = tokens['attention_mask'].to(device)
|
| 274 |
+
vector = model(input_ids, attention_mask)
|
| 275 |
+
generated_vectors.append(vector.cpu().detach().numpy().flatten())
|
| 276 |
+
|
| 277 |
+
df_filtrado['vector_estilo'] = generated_vectors
|
| 278 |
+
df_filtrado['vector_estilo'] = df_filtrado['vector_estilo'].apply(lambda x: np.array(x))
|
| 279 |
+
|
| 280 |
+
self.df_muebles = df_filtrado
|
| 281 |
+
# Guardar caché
|
| 282 |
+
try:
|
| 283 |
+
with open(self.cache_path, 'wb') as f:
|
| 284 |
+
pickle.dump(self.df_muebles, f)
|
| 285 |
+
except Exception as e:
|
| 286 |
+
print(f"Advertencia: No se pudo guardar la caché de vectores: {e}")
|
| 287 |
+
|
| 288 |
+
|
| 289 |
+
def _calcular_dimensiones_promedio(self):
|
| 290 |
+
"""Calcula el ancho y largo promedio por Tipo_mueble."""
|
| 291 |
+
if self.df_muebles is None: return
|
| 292 |
+
|
| 293 |
+
cols = ['Ancho', 'Largo', 'Altura']
|
| 294 |
+
for col in cols:
|
| 295 |
+
self.df_muebles[col] = pd.to_numeric(self.df_muebles[col], errors='coerce')
|
| 296 |
+
# Reemplazar valores <= 0 con NaN para no afectar el promedio
|
| 297 |
+
self.df_muebles.loc[self.df_muebles[col] <= 0, col] = np.nan
|
| 298 |
+
|
| 299 |
+
avg_df = self.df_muebles.groupby('Tipo_mueble')[cols].mean()
|
| 300 |
+
|
| 301 |
+
self.dimensiones_promedio = {}
|
| 302 |
+
for tipo, row in avg_df.iterrows():
|
| 303 |
+
self.dimensiones_promedio[tipo] = {
|
| 304 |
+
'nombre': tipo,
|
| 305 |
+
'ancho': round(row['Ancho'], 1) if not pd.isna(row['Ancho']) else 100.0,
|
| 306 |
+
'largo': round(row['Largo'], 1) if not pd.isna(row['Largo']) else 50.0
|
| 307 |
+
}
|
| 308 |
+
|
| 309 |
+
# Correcciones de lógica de negocio o datos faltantes (en cm)
|
| 310 |
+
if 'Sofás' in self.dimensiones_promedio:
|
| 311 |
+
if self.dimensiones_promedio['Sofás']['largo'] < 50: self.dimensiones_promedio['Sofás']['largo'] = 90.0
|
| 312 |
+
if 'Sillones' in self.dimensiones_promedio:
|
| 313 |
+
if self.dimensiones_promedio['Sillones']['largo'] < 40:
|
| 314 |
+
self.dimensiones_promedio['Sillones']['largo'] = self.dimensiones_promedio['Sillones']['ancho'] if self.dimensiones_promedio['Sillones']['ancho'] >= 40 else 70.0
|
| 315 |
+
|
| 316 |
+
# ==========================================
|
| 317 |
+
# 3. MOTOR DE LAYOUT (Lógica Espacial y de Colocación)
|
| 318 |
+
# ==========================================
|
| 319 |
+
class LayoutEngine:
|
| 320 |
+
"""Implementa la lógica de colocación de muebles, restricciones espaciales y colisiones."""
|
| 321 |
+
def __init__(self, dimensiones_promedio):
|
| 322 |
+
self.dim_promedio = dimensiones_promedio
|
| 323 |
+
self.config = {
|
| 324 |
+
'pasillo_minimo': 90, # cm
|
| 325 |
+
'margen_pared': 10, # cm
|
| 326 |
+
'margen_obstaculo': 20, # cm
|
| 327 |
+
'distancia_tv_sofa': 280 # cm (distancia ideal)
|
| 328 |
+
}
|
| 329 |
+
|
| 330 |
+
def sugerir_pack(self, ancho_cm, largo_cm):
|
| 331 |
+
"""Sugiere un conjunto de muebles base según el área de la habitación."""
|
| 332 |
+
area = (ancho_cm * largo_cm) / 10000.0
|
| 333 |
+
# Definición de tipos de mueble base
|
| 334 |
+
pack = [{'tipo': 'Muebles de salón'}, {'tipo': 'Mesas bajas de salón, de centro y auxiliares'}, {'tipo': 'Sofás'}]
|
| 335 |
+
# Añadir complejidad según el tamaño de la sala
|
| 336 |
+
if area > 16.0: pack.insert(0, {'tipo': 'Sillones'})
|
| 337 |
+
if area > 22.0: pack.insert(0, {'tipo': 'Estanterías y librerías'})
|
| 338 |
+
return pack
|
| 339 |
+
|
| 340 |
+
def convertir_obstaculos(self, obstaculos_dict, ancho_hab, largo_hab, polygon_points):
|
| 341 |
+
"""Convierte los obstáculos detectados (puertas/ventanas) en polígonos Shapely con margen de seguridad."""
|
| 342 |
+
obs_polys = []
|
| 343 |
+
num_walls = len(polygon_points)
|
| 344 |
+
|
| 345 |
+
todos_obs = []
|
| 346 |
+
for d in obstaculos_dict.get('doors', []): d['type'] = 'door'; todos_obs.append(d)
|
| 347 |
+
for w in obstaculos_dict.get('windows', []): w['type'] = 'window'; todos_obs.append(w)
|
| 348 |
+
|
| 349 |
+
for obs in todos_obs:
|
| 350 |
+
# Lógica para mapear la posición normalizada (0-1) a un polígono en CM
|
| 351 |
+
w_idx = int(round(obs['center'][1] * num_walls)) % num_walls
|
| 352 |
+
p1 = np.array(polygon_points[w_idx]) * 100
|
| 353 |
+
p2 = np.array(polygon_points[(w_idx + 1) % num_walls]) * 100
|
| 354 |
+
|
| 355 |
+
vec_pared = p2 - p1
|
| 356 |
+
len_pared = np.linalg.norm(vec_pared)
|
| 357 |
+
unit_pared = vec_pared / len_pared
|
| 358 |
+
|
| 359 |
+
# Centro del obstáculo a lo largo de la pared
|
| 360 |
+
center_pt = p1 + vec_pared * obs['center'][0]
|
| 361 |
+
width = obs['width'] * 100
|
| 362 |
+
depth = 250 if obs['type'] == 'door' else 30 # Profundidad de la zona de exclusión (250cm = ~1.25m dentro de la habitación)
|
| 363 |
+
|
| 364 |
+
# Vector normal a la pared
|
| 365 |
+
normal = np.array([-unit_pared[1], unit_pared[0]])
|
| 366 |
+
|
| 367 |
+
# Definir las esquinas del polígono de exclusión (con margen)
|
| 368 |
+
c1 = center_pt - unit_pared * (width/2) - normal * (depth/2)
|
| 369 |
+
c2 = center_pt + unit_pared * (width/2) - normal * (depth/2)
|
| 370 |
+
c3 = center_pt + unit_pared * (width/2) + normal * (depth/2)
|
| 371 |
+
c4 = center_pt - unit_pared * (width/2) + normal * (depth/2)
|
| 372 |
+
|
| 373 |
+
poly = Polygon([c1, c2, c3, c4])
|
| 374 |
+
obs_polys.append({'poly': poly, 'tipo': obs['type']})
|
| 375 |
+
|
| 376 |
+
return obs_polys
|
| 377 |
+
|
| 378 |
+
def _get_poly_from_rect(self, x, y, w, l, angle_rad):
|
| 379 |
+
"""Genera un Polygon de Shapely rotado a partir del centro (x, y), dimensiones (w, l) y ángulo."""
|
| 380 |
+
cx, cy = x, y
|
| 381 |
+
dx = w / 2
|
| 382 |
+
dy = l / 2
|
| 383 |
+
|
| 384 |
+
corners = [
|
| 385 |
+
(dx, dy), (-dx, dy), (-dx, -dy), (dx, -dy)
|
| 386 |
+
]
|
| 387 |
+
|
| 388 |
+
new_corners = []
|
| 389 |
+
c_cos = np.cos(angle_rad)
|
| 390 |
+
c_sin = np.sin(angle_rad)
|
| 391 |
+
|
| 392 |
+
for px, py in corners:
|
| 393 |
+
nx = px * c_cos - py * c_sin + cx
|
| 394 |
+
ny = px * c_sin + py * c_cos + cy
|
| 395 |
+
new_corners.append((nx, ny))
|
| 396 |
+
|
| 397 |
+
return Polygon(new_corners)
|
| 398 |
+
|
| 399 |
+
def _check_collision(self, candidate_poly, room_poly, placed_items, obstacles):
|
| 400 |
+
"""Verifica si el polígono candidato colisiona con la pared, ítems colocados u obstáculos."""
|
| 401 |
+
# 1. Dentro de la habitación (con margen de pared)
|
| 402 |
+
buffered_room = room_poly.buffer(-self.config['margen_pared'])
|
| 403 |
+
if not buffered_room.contains(candidate_poly):
|
| 404 |
+
# print(f"DEBUG: Colisión con PARED. Poly fuera del buffer.")
|
| 405 |
+
# print(f" Room Buffer Bounds: {buffered_room.bounds}")
|
| 406 |
+
# print(f" Candidate Bounds: {candidate_poly.bounds}")
|
| 407 |
+
return False
|
| 408 |
+
|
| 409 |
+
# 2. Colisión con items ya colocados
|
| 410 |
+
for item in placed_items:
|
| 411 |
+
# Usar un buffer de 10cm para asegurar un pequeño espacio entre muebles
|
| 412 |
+
if candidate_poly.buffer(10).intersects(item['poly']):
|
| 413 |
+
# print(f"DEBUG: Colisión con ITEM {item['tipo']}.")
|
| 414 |
+
return False
|
| 415 |
+
|
| 416 |
+
# 3. Colisión con obstáculos (con margen)
|
| 417 |
+
for obs in obstacles:
|
| 418 |
+
if candidate_poly.intersects(obs['poly']):
|
| 419 |
+
# print(f"DEBUG: Colisión con OBSTÁCULO {obs['tipo']}.")
|
| 420 |
+
return False
|
| 421 |
+
|
| 422 |
+
return True
|
| 423 |
+
|
| 424 |
+
def _scan_wall(self, p1, p2, item_dim, room_poly, placed, obstacles, align_dist=None, align_target=None):
|
| 425 |
+
"""Barre una pared específica buscando la primera ubicación válida para un mueble."""
|
| 426 |
+
vec = p2 - p1
|
| 427 |
+
wall_len = np.linalg.norm(vec)
|
| 428 |
+
unit_vec = vec / wall_len
|
| 429 |
+
|
| 430 |
+
normal = np.array([-unit_vec[1], unit_vec[0]])
|
| 431 |
+
|
| 432 |
+
# Verificar que la normal apunta al interior de la habitación
|
| 433 |
+
centroid = np.array(room_poly.centroid.coords[0])
|
| 434 |
+
mid_wall = (p1 + p2) / 2
|
| 435 |
+
if np.dot(centroid - mid_wall, normal) < 0:
|
| 436 |
+
normal = -normal
|
| 437 |
+
|
| 438 |
+
w_item = item_dim['ancho']
|
| 439 |
+
l_item = item_dim['largo']
|
| 440 |
+
angle = np.arctan2(unit_vec[1], unit_vec[0])
|
| 441 |
+
|
| 442 |
+
# Distancia del centro del mueble a la pared (para pegar a la pared)
|
| 443 |
+
dist_from_wall = self.config['margen_pared'] + l_item/2
|
| 444 |
+
if align_dist: dist_from_wall = align_dist
|
| 445 |
+
|
| 446 |
+
step = 20 # cm
|
| 447 |
+
margin_side = self.config['margen_pared'] + w_item/2
|
| 448 |
+
|
| 449 |
+
range_start = margin_side
|
| 450 |
+
range_end = wall_len - margin_side
|
| 451 |
+
|
| 452 |
+
if range_end < range_start: return []
|
| 453 |
+
|
| 454 |
+
candidates = list(np.arange(range_start, range_end, step))
|
| 455 |
+
# Asegurar que el centro de la pared está en los candidatos
|
| 456 |
+
mid_dist = wall_len / 2
|
| 457 |
+
if range_start <= mid_dist <= range_end:
|
| 458 |
+
candidates.append(mid_dist)
|
| 459 |
+
# Ordenar para probar primero el centro? No necesariamente, pero ayuda.
|
| 460 |
+
candidates = sorted(list(set(candidates)))
|
| 461 |
+
|
| 462 |
+
if align_target is not None:
|
| 463 |
+
# Lógica para alinear con un objeto existente
|
| 464 |
+
v_target = np.array([align_target['x'], align_target['y']]) - p1
|
| 465 |
+
proj_dist = np.dot(v_target, unit_vec)
|
| 466 |
+
candidates = [proj_dist]
|
| 467 |
+
|
| 468 |
+
valid_candidates = []
|
| 469 |
+
for dist_along in candidates:
|
| 470 |
+
center = p1 + unit_vec * dist_along + normal * dist_from_wall
|
| 471 |
+
cand_poly = self._get_poly_from_rect(center[0], center[1], w_item, l_item, angle)
|
| 472 |
+
|
| 473 |
+
if self._check_collision(cand_poly, room_poly, placed, obstacles):
|
| 474 |
+
valid_candidates.append({
|
| 475 |
+
'x': center[0], 'y': center[1],
|
| 476 |
+
'ancho': w_item, 'largo': l_item,
|
| 477 |
+
'angle': angle,
|
| 478 |
+
'tipo': item_dim['nombre'],
|
| 479 |
+
'poly': cand_poly
|
| 480 |
+
})
|
| 481 |
+
return valid_candidates
|
| 482 |
+
|
| 483 |
+
def generar_layout(self, ancho_hab, largo_hab, pack_sugerido, obs_layout, polygon_points=None):
|
| 484 |
+
"""Genera el layout buscando una configuración TV-Sofá válida y añadiendo la mesa de centro."""
|
| 485 |
+
# setup básico
|
| 486 |
+
layout = []
|
| 487 |
+
constraints = []
|
| 488 |
+
log = []
|
| 489 |
+
final_layout_objs = []
|
| 490 |
+
|
| 491 |
+
if not polygon_points:
|
| 492 |
+
# Usar un rectángulo simple si no hay polígono
|
| 493 |
+
polygon_points = [[0,0], [ancho_hab/100, 0], [ancho_hab/100, largo_hab/100], [0, largo_hab/100]]
|
| 494 |
+
|
| 495 |
+
poly_pts_cm = np.array(polygon_points) * 100
|
| 496 |
+
room_poly = Polygon(poly_pts_cm)
|
| 497 |
+
num_walls = len(poly_pts_cm)
|
| 498 |
+
|
| 499 |
+
# Dimensiones promedio de los dos muebles clave
|
| 500 |
+
d_tv = self.dim_promedio.get('Muebles de salón', {'ancho': 120, 'largo': 40})
|
| 501 |
+
d_tv['nombre'] = 'Muebles de salón'
|
| 502 |
+
d_sofa = self.dim_promedio.get('Sofás', {'ancho': 200, 'largo': 90})
|
| 503 |
+
d_sofa['nombre'] = 'Sofás'
|
| 504 |
+
|
| 505 |
+
# Estrategia de reintento con tamaños reducidos
|
| 506 |
+
attempt_configs = [
|
| 507 |
+
{'scale': 1.0, 'desc': 'Standard'},
|
| 508 |
+
{'scale': 0.8, 'desc': 'Compact'}
|
| 509 |
+
]
|
| 510 |
+
|
| 511 |
+
best_overall_score = -1
|
| 512 |
+
best_overall_layout = []
|
| 513 |
+
|
| 514 |
+
for config in attempt_configs:
|
| 515 |
+
scale = config['scale']
|
| 516 |
+
# Aplicar escala a las dimensiones temporales para este intento
|
| 517 |
+
current_d_tv = d_tv.copy()
|
| 518 |
+
current_d_tv['ancho'] *= scale
|
| 519 |
+
current_d_tv['largo'] *= scale
|
| 520 |
+
|
| 521 |
+
current_d_sofa = d_sofa.copy()
|
| 522 |
+
current_d_sofa['ancho'] *= scale
|
| 523 |
+
current_d_sofa['largo'] *= scale
|
| 524 |
+
|
| 525 |
+
log.append(f"🔄 Intentando generación con modo {config['desc']} (Escala {scale})")
|
| 526 |
+
|
| 527 |
+
best_score = -1
|
| 528 |
+
best_layout = []
|
| 529 |
+
|
| 530 |
+
# Iterar sobre todas las paredes para encontrar la mejor ubicación para el binomio TV-Sofá
|
| 531 |
+
for i in range(num_walls):
|
| 532 |
+
p1 = poly_pts_cm[i]
|
| 533 |
+
p2 = poly_pts_cm[(i+1)%num_walls]
|
| 534 |
+
|
| 535 |
+
# 1. Intentar poner Mueble de Salón (TV) en la pared
|
| 536 |
+
# Ahora devuelve una lista de candidatos
|
| 537 |
+
tv_candidates = self._scan_wall(p1, p2, current_d_tv, room_poly, [], obs_layout)
|
| 538 |
+
|
| 539 |
+
for tv_pos in tv_candidates:
|
| 540 |
+
current_layout = [tv_pos]
|
| 541 |
+
|
| 542 |
+
# Calcular la normal del TV
|
| 543 |
+
vec_pared = p2 - p1
|
| 544 |
+
unit_vec_pared = vec_pared / np.linalg.norm(vec_pared)
|
| 545 |
+
normal_base = np.array([-unit_vec_pared[1], unit_vec_pared[0]])
|
| 546 |
+
|
| 547 |
+
centroid = np.array(room_poly.centroid.coords[0])
|
| 548 |
+
mid_wall = (p1 + p2) / 2
|
| 549 |
+
|
| 550 |
+
# Asegurar que la normal apunta al centro (adentro)
|
| 551 |
+
if np.linalg.norm((mid_wall + normal_base) - centroid) > np.linalg.norm((mid_wall - normal_base) - centroid):
|
| 552 |
+
tv_normal = -normal_base
|
| 553 |
+
else:
|
| 554 |
+
tv_normal = normal_base
|
| 555 |
+
|
| 556 |
+
# 2. Calcular dónde estaría el sofá y proyectar un rayo
|
| 557 |
+
tv_center_pt = Point(tv_pos['x'], tv_pos['y'])
|
| 558 |
+
ray_end_np = np.array([tv_pos['x'], tv_pos['y']]) + tv_normal * max(ancho_hab, largo_hab) * 100
|
| 559 |
+
ray = LineString([tv_center_pt, (ray_end_np[0], ray_end_np[1])])
|
| 560 |
+
|
| 561 |
+
intersection = ray.intersection(room_poly.boundary)
|
| 562 |
+
|
| 563 |
+
distancia_pared_opuesta = 9999
|
| 564 |
+
|
| 565 |
+
if not intersection.is_empty:
|
| 566 |
+
if intersection.geom_type == 'Point':
|
| 567 |
+
d = tv_center_pt.distance(intersection)
|
| 568 |
+
if d > 50: distancia_pared_opuesta = d
|
| 569 |
+
elif intersection.geom_type == 'MultiPoint':
|
| 570 |
+
for pt in intersection.geoms:
|
| 571 |
+
d = tv_center_pt.distance(pt)
|
| 572 |
+
if d > 50 and d < distancia_pared_opuesta:
|
| 573 |
+
distancia_pared_opuesta = d
|
| 574 |
+
|
| 575 |
+
dist_ideal = self.config['distancia_tv_sofa']
|
| 576 |
+
fondo_sofa = current_d_sofa['largo']
|
| 577 |
+
|
| 578 |
+
# Calcular el espacio libre entre la parte trasera del sofá y la pared
|
| 579 |
+
espacio_detras = distancia_pared_opuesta - (tv_pos['largo']/2 + dist_ideal + fondo_sofa/2)
|
| 580 |
+
|
| 581 |
+
dist_sofa_desde_tv = dist_ideal
|
| 582 |
+
|
| 583 |
+
if distancia_pared_opuesta < (tv_pos['largo']/2 + dist_ideal + fondo_sofa + self.config['margen_pared']):
|
| 584 |
+
# Si la habitación es demasiado pequeña, pega el sofá a la pared trasera
|
| 585 |
+
# Corrección: No restar tv_pos['largo']/2 porque la distancia es desde el centro de la TV
|
| 586 |
+
dist_sofa_desde_tv = distancia_pared_opuesta - fondo_sofa/2 - self.config['margen_pared']
|
| 587 |
+
log_msg = f"✅ Sofá ajustado a pared (Espacio insuficiente para ideal)"
|
| 588 |
+
elif espacio_detras < 100: # Aumentado a 100cm. Si sobra menos de 1m, pégalo atrás.
|
| 589 |
+
# Corrección: Añadir 5cm extra de margen para asegurar que el polígono esté DENTRO del buffer de la habitación
|
| 590 |
+
dist_sofa_desde_tv = distancia_pared_opuesta - fondo_sofa/2 - self.config['margen_pared'] - 5.0
|
| 591 |
+
log_msg = f"✅ Sofá ajustado a pared (Evitar espacio muerto de {espacio_detras:.0f}cm)"
|
| 592 |
+
else:
|
| 593 |
+
# Posicionamiento ideal frente a TV
|
| 594 |
+
dist_sofa_desde_tv = dist_ideal + tv_pos['largo']/2 + fondo_sofa/2
|
| 595 |
+
log_msg = "✅ Sofá en isla"
|
| 596 |
+
|
| 597 |
+
|
| 598 |
+
# Intentar colocar el sofá con "nudge" (empujoncitos) SOLO frontal para mantener alineación
|
| 599 |
+
sofa_placed = False
|
| 600 |
+
# Nudge frontal: 0 a 30cm
|
| 601 |
+
for nudge in [0, 5, 10, 15, 20, 25, 30]:
|
| 602 |
+
current_dist = dist_sofa_desde_tv - nudge
|
| 603 |
+
|
| 604 |
+
sofa_center_np = np.array([tv_pos['x'], tv_pos['y']]) + tv_normal * current_dist
|
| 605 |
+
sofa_poly = self._get_poly_from_rect(sofa_center_np[0], sofa_center_np[1], current_d_sofa['ancho'], current_d_sofa['largo'], tv_pos['angle'])
|
| 606 |
+
|
| 607 |
+
sofa_cand = {
|
| 608 |
+
'x': sofa_center_np[0], 'y': sofa_center_np[1],
|
| 609 |
+
'ancho': current_d_sofa['ancho'], 'largo': current_d_sofa['largo'],
|
| 610 |
+
'angle': tv_pos['angle'], 'tipo': 'Sofás', 'poly': sofa_poly
|
| 611 |
+
}
|
| 612 |
+
|
| 613 |
+
if self._check_collision(sofa_poly, room_poly, [tv_pos], obs_layout):
|
| 614 |
+
current_layout.append(sofa_cand)
|
| 615 |
+
score = 100 - nudge
|
| 616 |
+
if score > best_score:
|
| 617 |
+
best_score = score
|
| 618 |
+
best_layout = current_layout
|
| 619 |
+
log.append(f"{log_msg} (Nudge={nudge}cm)")
|
| 620 |
+
sofa_placed = True
|
| 621 |
+
break
|
| 622 |
+
|
| 623 |
+
if sofa_placed:
|
| 624 |
+
break # Romper el loop de TV candidates si ya encontramos un layout válido
|
| 625 |
+
else:
|
| 626 |
+
log.append(f"❌ Sofá colisiona tras intentos. DistBase={dist_sofa_desde_tv:.1f}")
|
| 627 |
+
# print(f"DEBUG: Fallo Sofá. Dist={dist_sofa_desde_tv:.1f}. Wall={i}")
|
| 628 |
+
|
| 629 |
+
if best_layout:
|
| 630 |
+
best_overall_layout = best_layout
|
| 631 |
+
best_overall_score = best_score
|
| 632 |
+
break # Si encontramos un layout válido con esta escala, nos quedamos con él
|
| 633 |
+
|
| 634 |
+
best_layout = best_overall_layout # Restaurar para el resto del código
|
| 635 |
+
|
| 636 |
+
if best_layout:
|
| 637 |
+
# 2. Convertir layout a formato final y generar restricciones
|
| 638 |
+
for item in best_layout:
|
| 639 |
+
final_layout_objs.append({
|
| 640 |
+
'x': item['x'], 'y': item['y'],
|
| 641 |
+
'ancho': item['ancho'], 'largo': item['largo'],
|
| 642 |
+
'angle': item['angle'], 'tipo': item['tipo']
|
| 643 |
+
})
|
| 644 |
+
# Definir la restricción dimensional
|
| 645 |
+
constraints.append({'tipo': item['tipo'], 'max_ancho': item['ancho']*1.2, 'max_largo': item['largo']*1.2})
|
| 646 |
+
|
| 647 |
+
# 3. Colocación condicional de Mesa de Centro
|
| 648 |
+
tv = best_layout[0]
|
| 649 |
+
sofa = best_layout[1]
|
| 650 |
+
|
| 651 |
+
# Calcular la distancia libre entre TV y Sofá
|
| 652 |
+
dist_centros = np.linalg.norm(np.array([tv['x'], tv['y']]) - np.array([sofa['x'], sofa['y']]))
|
| 653 |
+
depth_tv = tv['largo']
|
| 654 |
+
depth_sofa = sofa['largo']
|
| 655 |
+
|
| 656 |
+
espacio_libre = dist_centros - (depth_tv / 2) - (depth_sofa / 2)
|
| 657 |
+
|
| 658 |
+
profundidad_mesa = self.dim_promedio.get('Mesas bajas de salón, de centro y auxiliares', {'largo': 60})['largo']
|
| 659 |
+
pasillo_minimo = 60 # cm para circular alrededor
|
| 660 |
+
|
| 661 |
+
if espacio_libre >= (profundidad_mesa + pasillo_minimo):
|
| 662 |
+
mid_x = (tv['x'] + sofa['x']) / 2
|
| 663 |
+
mid_y = (tv['y'] + sofa['y']) / 2
|
| 664 |
+
|
| 665 |
+
final_layout_objs.append({
|
| 666 |
+
'x': mid_x, 'y': mid_y,
|
| 667 |
+
'ancho': 100, 'largo': profundidad_mesa, # Usamos ancho y largo genérico
|
| 668 |
+
'angle': tv['angle'],
|
| 669 |
+
'tipo': 'Mesas bajas de salón, de centro y auxiliares'
|
| 670 |
+
})
|
| 671 |
+
constraints.append({'tipo': 'Mesas bajas de salón, de centro y auxiliares'})
|
| 672 |
+
else:
|
| 673 |
+
log.append(f"⚠️ Mesa de centro omitida: Espacio libre ({espacio_libre:.0f}cm) insuficiente para mesa + paso.")
|
| 674 |
+
|
| 675 |
+
else:
|
| 676 |
+
log.append("No se encontró distribución válida TV-Sofá. Distribución fallida.")
|
| 677 |
+
|
| 678 |
+
return final_layout_objs, constraints, log
|
| 679 |
+
|
| 680 |
+
# ==========================================
|
| 681 |
+
# 4. MOTOR DE RECOMENDACIÓN (Knapsack/Estilo)
|
| 682 |
+
# ==========================================
|
| 683 |
+
class Recommender:
|
| 684 |
+
"""Implementa el algoritmo de selección de productos para maximizar el Score/Coherencia dentro del presupuesto."""
|
| 685 |
+
def __init__(self, df_data):
|
| 686 |
+
self.df = df_data
|
| 687 |
+
|
| 688 |
+
def _coherencia(self, vectores):
|
| 689 |
+
"""Calcula la coherencia de estilo promedio (Similitud Coseno) entre los vectores de estilo de los muebles."""
|
| 690 |
+
if len(vectores) < 2: return 1.0
|
| 691 |
+
mat = cosine_similarity(np.array(vectores))
|
| 692 |
+
# Promedio de la similitud entre todos los pares (triángulo superior)
|
| 693 |
+
indices = np.triu_indices_from(mat, k=1)
|
| 694 |
+
return float(np.mean(mat[indices])) if indices[0].size > 0 else 1.0
|
| 695 |
+
|
| 696 |
+
def buscar_combinacion(self, constraints, presupuesto, top_n=1):
|
| 697 |
+
"""
|
| 698 |
+
Algoritmo Knapsack de fuerza bruta optimizada.
|
| 699 |
+
Busca la mejor combinación de productos que cumpla restricciones dimensionales y presupuestarias,
|
| 700 |
+
maximizando el score total (Score Base + Coherencia de Estilo).
|
| 701 |
+
"""
|
| 702 |
+
listas_candidatos = []
|
| 703 |
+
|
| 704 |
+
for const in constraints:
|
| 705 |
+
tipo = const['tipo']
|
| 706 |
+
max_w = const.get('max_ancho', 9999)
|
| 707 |
+
max_l = const.get('max_largo', 9999)
|
| 708 |
+
|
| 709 |
+
# Filtro dimensional
|
| 710 |
+
pool = self.df[self.df['Tipo_mueble'] == tipo]
|
| 711 |
+
# Permite rotación (Ancho <= Max_W y Largo <= Max_L) o (Ancho <= Max_L y Largo <= Max_W)
|
| 712 |
+
fits = pool[((pool['Ancho'] <= max_w) & (pool['Largo'] <= max_l)) |
|
| 713 |
+
((pool['Ancho'] <= max_l) & (pool['Largo'] <= max_w))]
|
| 714 |
+
|
| 715 |
+
if fits.empty: fits = pool # Si no hay que cumplen, toma la lista completa
|
| 716 |
+
|
| 717 |
+
# ESTRATEGIA HÍBRIDA:
|
| 718 |
+
# Seleccionar los top 5 por Score (Calidad) Y los top 5 más baratos (Presupuesto)
|
| 719 |
+
# para asegurar que tenemos opciones viables si el presupuesto es bajo.
|
| 720 |
+
|
| 721 |
+
top_score = fits.sort_values('Score', ascending=False).head(5)
|
| 722 |
+
top_cheap = fits.sort_values('Precio', ascending=True).head(5)
|
| 723 |
+
top_expensive = fits.sort_values('Precio', ascending=False).head(5)
|
| 724 |
+
|
| 725 |
+
# Combinar y eliminar duplicados (usando ID para evitar error con numpy arrays)
|
| 726 |
+
candidates_df = pd.concat([top_score, top_cheap, top_expensive]).drop_duplicates(subset='ID')
|
| 727 |
+
|
| 728 |
+
candidatos = candidates_df.to_dict('records')
|
| 729 |
+
listas_candidatos.append(candidatos)
|
| 730 |
+
|
| 731 |
+
if not listas_candidatos: return []
|
| 732 |
+
|
| 733 |
+
validas = []
|
| 734 |
+
for combo in itertools.product(*listas_candidatos):
|
| 735 |
+
precio = sum(x['Precio'] for x in combo)
|
| 736 |
+
if precio <= presupuesto:
|
| 737 |
+
score_base = np.mean([x['Score'] for x in combo])
|
| 738 |
+
vectores = [x['vector_estilo'] for x in combo]
|
| 739 |
+
coherencia = self._coherencia(vectores)
|
| 740 |
+
|
| 741 |
+
# Score Final:
|
| 742 |
+
# 40% Score Base (Calidad/Popularidad)
|
| 743 |
+
# 40% Coherencia (Estilo)
|
| 744 |
+
# 20% Aprovechamiento de Presupuesto (Reward por usar el presupuesto disponible)
|
| 745 |
+
budget_utilization = precio / presupuesto
|
| 746 |
+
|
| 747 |
+
final_score = 0.4 * score_base + 0.4 * coherencia + 0.2 * budget_utilization
|
| 748 |
+
|
| 749 |
+
validas.append({
|
| 750 |
+
'items': combo,
|
| 751 |
+
'precio_total': precio,
|
| 752 |
+
'score': final_score
|
| 753 |
+
})
|
| 754 |
+
|
| 755 |
+
validas.sort(key=lambda x: x['score'], reverse=True)
|
| 756 |
+
return validas[:top_n]
|
| 757 |
+
|
| 758 |
+
# ==========================================
|
| 759 |
+
# 5. VISUALIZADORES (PLANTAS 2D y 3D con OBJs)
|
| 760 |
+
# ==========================================
|
| 761 |
+
|
| 762 |
+
def get_segment_properties(p1, p2):
|
| 763 |
+
"""Calcula longitud, punto central y ángulo de un segmento 2D."""
|
| 764 |
+
p1 = np.array(p1)
|
| 765 |
+
p2 = np.array(p2)
|
| 766 |
+
|
| 767 |
+
dx = p2[0] - p1[0]
|
| 768 |
+
dy = p2[1] - p1[1]
|
| 769 |
+
length = np.sqrt(dx**2 + dy**2)
|
| 770 |
+
midpoint_x = (p1[0] + p2[0]) / 2
|
| 771 |
+
midpoint_y = (p1[1] + p2[1]) / 2
|
| 772 |
+
angle = np.arctan2(dy, dx)
|
| 773 |
+
return length, midpoint_x, midpoint_y, angle
|
| 774 |
+
|
| 775 |
+
|
| 776 |
+
def read_kenney_obj(obj_path):
|
| 777 |
+
"""
|
| 778 |
+
Lee archivos OBJ simples (como los de Kenney) extrayendo solo vértices (v) y caras (f).
|
| 779 |
+
Retorna (vertices_list, faces_list).
|
| 780 |
+
"""
|
| 781 |
+
vertices = []
|
| 782 |
+
faces = []
|
| 783 |
+
|
| 784 |
+
# --- DEBUG: Comprobar lectura de archivo ---
|
| 785 |
+
print(f"\n--- DEBUG: Leyendo manualmente: {obj_path} ---")
|
| 786 |
+
v_count = 0
|
| 787 |
+
f_count = 0
|
| 788 |
+
|
| 789 |
+
try:
|
| 790 |
+
# Nota: El error podría ser la codificación. Usamos 'utf-8' o 'latin-1'
|
| 791 |
+
with open(obj_path, 'r', encoding='latin-1') as f:
|
| 792 |
+
for line in f:
|
| 793 |
+
parts = line.strip().split()
|
| 794 |
+
if not parts:
|
| 795 |
+
continue
|
| 796 |
+
|
| 797 |
+
prefix = parts[0]
|
| 798 |
+
|
| 799 |
+
if prefix == 'v':
|
| 800 |
+
# Vértices: 'v x y z'
|
| 801 |
+
try:
|
| 802 |
+
vertices.append([float(parts[1]), float(parts[2]), float(parts[3])])
|
| 803 |
+
v_count += 1
|
| 804 |
+
except ValueError:
|
| 805 |
+
print(f"DEBUG: Vértice inválido en {obj_name}: {line.strip()}")
|
| 806 |
+
|
| 807 |
+
elif prefix == 'f':
|
| 808 |
+
# Caras: 'f v/vt/vn v/vt/vn v/vt/vn ...'
|
| 809 |
+
try:
|
| 810 |
+
face_indices = []
|
| 811 |
+
for part in parts[1:]:
|
| 812 |
+
v_index = int(part.split('/')[0])
|
| 813 |
+
face_indices.append(v_index - 1)
|
| 814 |
+
faces.append(face_indices)
|
| 815 |
+
f_count += 1
|
| 816 |
+
except ValueError:
|
| 817 |
+
print(f"DEBUG: Cara inválida en {obj_name}: {line.strip()}")
|
| 818 |
+
|
| 819 |
+
except FileNotFoundError:
|
| 820 |
+
print(f"DEBUG: Archivo no encontrado en la ruta de {obj_path}")
|
| 821 |
+
return [], []
|
| 822 |
+
except Exception as e:
|
| 823 |
+
print(f"DEBUG: Error inesperado de E/S: {e}")
|
| 824 |
+
return [], []
|
| 825 |
+
|
| 826 |
+
# --- DEBUG: Reporte final de la lectura ---
|
| 827 |
+
print(f"DEBUG RESULTADO: Vértices leídos (v): {v_count}")
|
| 828 |
+
print(f"DEBUG RESULTADO: Caras leídas (f): {f_count}")
|
| 829 |
+
print(f"---------------------------------------------")
|
| 830 |
+
|
| 831 |
+
return vertices, faces
|
| 832 |
+
|
| 833 |
+
|
| 834 |
+
def load_and_transform_mesh(obj_name, w, l, h, cx, cy, angle, base_z=0, rotation_offset=0):
|
| 835 |
+
"""
|
| 836 |
+
Carga un modelo .obj usando el parser manual, lo escala de forma NO UNIFORME
|
| 837 |
+
para encajar en (w, l, h), y lo rota/traslada a la posición.
|
| 838 |
+
"""
|
| 839 |
+
obj_path = os.path.join(MODEL_DIR, obj_name)
|
| 840 |
+
|
| 841 |
+
if not os.path.exists(obj_path):
|
| 842 |
+
print(f"!!! ERROR MODELO 3D: '{obj_name}' no encontrado en {MODEL_DIR}")
|
| 843 |
+
return None, None, None, None, None, None
|
| 844 |
+
|
| 845 |
+
# --- 1. CARGA USANDO PARSER MANUAL ---
|
| 846 |
+
vertices, faces_indices_list = read_kenney_obj(obj_path)
|
| 847 |
+
|
| 848 |
+
if not vertices:
|
| 849 |
+
print(f"!!! ERROR MODELO 3D: Modelo '{obj_name}' sin datos 3D después del parseo.")
|
| 850 |
+
return None, None, None, None, None, None
|
| 851 |
+
|
| 852 |
+
vertices_np = np.array(vertices, dtype=np.float32).reshape(-1, 3)
|
| 853 |
+
|
| 854 |
+
# --- 2. PREPARAR CARAS PARA PLOTLY ---
|
| 855 |
+
i_faces, j_faces, k_faces = [], [], []
|
| 856 |
+
|
| 857 |
+
for face in faces_indices_list:
|
| 858 |
+
if len(face) == 3:
|
| 859 |
+
i_faces.append(face[0])
|
| 860 |
+
j_faces.append(face[1])
|
| 861 |
+
k_faces.append(face[2])
|
| 862 |
+
elif len(face) == 4:
|
| 863 |
+
i_faces.extend([face[0], face[0]])
|
| 864 |
+
j_faces.extend([face[1], face[2]])
|
| 865 |
+
k_faces.extend([face[2], face[3]])
|
| 866 |
+
|
| 867 |
+
if not i_faces:
|
| 868 |
+
print(f"!!! ERROR MODELO 3D: Modelo '{obj_name}' sin caras válidas para Plotly.")
|
| 869 |
+
return None, None, None, None, None, None
|
| 870 |
+
|
| 871 |
+
# --- 3. TRANSFORMAR VÉRTICES (ESCALADO NO UNIFORME) ---
|
| 872 |
+
|
| 873 |
+
# 3.0. CORRECCIÓN CRÍTICA DE ORIENTACIÓN (Rotación 90° sobre X)
|
| 874 |
+
# Rota el modelo de Kenney (que suele tener Y=Arriba, Z=Profundidad) a
|
| 875 |
+
# la convención de tu sistema (Z=Arriba, Y=Profundidad).
|
| 876 |
+
|
| 877 |
+
# Matriz de rotación 90° sobre eje X (Rotar Y a Z, Z a -Y)
|
| 878 |
+
R_X = np.array([
|
| 879 |
+
[1, 0, 0],
|
| 880 |
+
[0, 0, -1],
|
| 881 |
+
[0, 1, 0]
|
| 882 |
+
])
|
| 883 |
+
vertices_np = vertices_np @ R_X.T # Aplicamos la rotación BASE
|
| 884 |
+
|
| 885 |
+
# 3.1. Encontrar el Bounding Box (para escalado) - ¡Usando los vértices rotados!
|
| 886 |
+
min_x, max_x = vertices_np[:, 0].min(), vertices_np[:, 0].max()
|
| 887 |
+
min_y, max_y = vertices_np[:, 1].min(), vertices_np[:, 1].max()
|
| 888 |
+
min_z, max_z = vertices_np[:, 2].min(), vertices_np[:, 2].max()
|
| 889 |
+
|
| 890 |
+
bbox_w = max_x - min_x
|
| 891 |
+
bbox_l = max_y - min_y
|
| 892 |
+
bbox_h = max_z - min_z
|
| 893 |
+
|
| 894 |
+
# 3.2. Calcular Factor de Escala NO UNIFORME
|
| 895 |
+
# Evitar división por cero
|
| 896 |
+
scale_x = w / bbox_w if bbox_w > 0 else 1.0
|
| 897 |
+
scale_y = l / bbox_l if bbox_l > 0 else 1.0
|
| 898 |
+
scale_z = h / bbox_h if bbox_h > 0 else 1.0
|
| 899 |
+
|
| 900 |
+
# 3.3. Trasladar al origen (centrar en XY y base en Z=0)
|
| 901 |
+
center_x_base = (min_x + max_x) / 2
|
| 902 |
+
center_y_base = (min_y + max_y) / 2
|
| 903 |
+
|
| 904 |
+
# Trasladar el centro y mover la base al plano Z=0
|
| 905 |
+
transformed_v = vertices_np - np.array([center_x_base, center_y_base, min_z])
|
| 906 |
+
|
| 907 |
+
# 3.4. Aplicar Escala No Uniforme
|
| 908 |
+
transformed_v[:, 0] *= scale_x
|
| 909 |
+
transformed_v[:, 1] *= scale_y
|
| 910 |
+
transformed_v[:, 2] *= scale_z
|
| 911 |
+
|
| 912 |
+
# --- 4. APLICAR ROTACIÓN Y TRASLACIÓN FINAL ---
|
| 913 |
+
|
| 914 |
+
# Aplicar Rotación del Layout (en el eje Z) + Offset
|
| 915 |
+
final_angle = angle + rotation_offset
|
| 916 |
+
c_cos = np.cos(final_angle)
|
| 917 |
+
c_sin = np.sin(final_angle)
|
| 918 |
+
rot_matrix = np.array([[c_cos, -c_sin], [c_sin, c_cos]])
|
| 919 |
+
transformed_v[:, :2] = transformed_v[:, :2] @ rot_matrix.T
|
| 920 |
+
|
| 921 |
+
# Aplicar Traslación Final
|
| 922 |
+
transformed_v[:, 0] += cx
|
| 923 |
+
transformed_v[:, 1] += cy
|
| 924 |
+
transformed_v[:, 2] += base_z
|
| 925 |
+
|
| 926 |
+
x_coords, y_coords, z_coords = transformed_v[:, 0], transformed_v[:, 1], transformed_v[:, 2]
|
| 927 |
+
|
| 928 |
+
return x_coords, y_coords, z_coords, i_faces, j_faces, k_faces
|
| 929 |
+
|
| 930 |
+
def dibujar_layout_sobre_imagen(img_path, room_data):
|
| 931 |
+
"""
|
| 932 |
+
Dibuja las predicciones de HorizonNet (líneas de floor/ceiling y corners)
|
| 933 |
+
sobre la imagen panorámica para visualización de la detección.
|
| 934 |
+
"""
|
| 935 |
+
try:
|
| 936 |
+
# Cargar imagen y redimensionar a 1024x512
|
| 937 |
+
img = Image.open(img_path).convert("RGB")
|
| 938 |
+
img = img.resize((1024, 512), Image.LANCZOS)
|
| 939 |
+
img_array = np.array(img)
|
| 940 |
+
|
| 941 |
+
# Verificar que tenemos los datos raw del modelo
|
| 942 |
+
if 'y_bon' not in room_data or 'y_cor' not in room_data:
|
| 943 |
+
draw = ImageDraw.Draw(img)
|
| 944 |
+
draw.text((10, 10), "Sin datos de visualización raw", fill=(255, 0, 0))
|
| 945 |
+
return img
|
| 946 |
+
|
| 947 |
+
y_bon = room_data['y_bon'] # [2, 1024] en radianes
|
| 948 |
+
y_cor = room_data['y_cor'] # [1024] probabilidades [0,1]
|
| 949 |
+
|
| 950 |
+
# Convertir boundary de radianes a píxeles
|
| 951 |
+
y_bon_pix = ((y_bon / np.pi + 0.5) * 512).round().astype(int)
|
| 952 |
+
y_bon_pix[0] = np.clip(y_bon_pix[0], 0, 511) # ceiling
|
| 953 |
+
y_bon_pix[1] = np.clip(y_bon_pix[1], 0, 511) # floor
|
| 954 |
+
|
| 955 |
+
# Crear visualización
|
| 956 |
+
img_vis = (img_array * 0.5).astype(np.uint8) # Oscurecer imagen un poco
|
| 957 |
+
|
| 958 |
+
# Dibujar las líneas de ceiling y floor (verde)
|
| 959 |
+
for x in range(1024):
|
| 960 |
+
img_vis[y_bon_pix[0][x], x] = [0, 255, 0] # Verde para ceiling
|
| 961 |
+
img_vis[y_bon_pix[1][x], x] = [0, 255, 0] # Verde para floor
|
| 962 |
+
|
| 963 |
+
# Dibujar probabilidades de corner como barra en la parte superior
|
| 964 |
+
cor_height = 30
|
| 965 |
+
gt_cor = np.zeros((cor_height, 1024, 3), np.uint8)
|
| 966 |
+
gt_cor[:] = (y_cor[None, :, None] * 255).astype(np.uint8) # Escala de grises
|
| 967 |
+
|
| 968 |
+
separator = np.ones((3, 1024, 3), np.uint8) * 255
|
| 969 |
+
|
| 970 |
+
# Concatenar: corner heatmap + separador + imagen con boundaries
|
| 971 |
+
final_vis = np.concatenate([gt_cor, separator, img_vis], axis=0)
|
| 972 |
+
|
| 973 |
+
return Image.fromarray(final_vis)
|
| 974 |
+
|
| 975 |
+
except Exception as e:
|
| 976 |
+
print(f"Error al dibujar layout: {e}")
|
| 977 |
+
import traceback
|
| 978 |
+
traceback.print_exc()
|
| 979 |
+
# En caso de error, devuelve la imagen original
|
| 980 |
+
return Image.open(img_path).convert("RGB")
|
| 981 |
+
|
| 982 |
+
|
| 983 |
+
def generar_figura_3d_plotly(layout_plan, room_data, items_recomendados, altura_pared=250):
|
| 984 |
+
"""
|
| 985 |
+
Genera visualización 3D interactiva, usando modelos .OBJ para estructura y muebles.
|
| 986 |
+
Si un OBJ no carga, el elemento es OMITIDO.
|
| 987 |
+
"""
|
| 988 |
+
if not layout_plan:
|
| 989 |
+
return go.Figure()
|
| 990 |
+
|
| 991 |
+
# --- Mapeo y Constantes ---
|
| 992 |
+
WALL_COLOR = '#bdbdbd'
|
| 993 |
+
CORNER_COLOR = '#6d6d6d'
|
| 994 |
+
DOOR_COLOR = '#8d6e63'
|
| 995 |
+
WINDOW_COLOR = '#d4e6f1'
|
| 996 |
+
WALL_H = altura_pared
|
| 997 |
+
WALL_DEPTH_CM = 10 # Profundidad de la pared a renderizar en 3D
|
| 998 |
+
|
| 999 |
+
MODEL_MAP = {
|
| 1000 |
+
'Sofás': 'loungeSofa.obj',
|
| 1001 |
+
'Sillones': 'loungeChair.obj',
|
| 1002 |
+
'Mesas bajas de salón, de centro y auxiliares': 'tableCoffee.obj',
|
| 1003 |
+
'Muebles de salón': 'cabinetTelevision.obj',
|
| 1004 |
+
'Estanterías y librerías': 'bookcaseOpen.obj'
|
| 1005 |
+
}
|
| 1006 |
+
|
| 1007 |
+
OBSTACLE_MAP = {
|
| 1008 |
+
# w y l representan las dimensiones del OBJ
|
| 1009 |
+
'door': {'obj': 'wallDoorway.obj', 'color': DOOR_COLOR, 'height': 200, 'v_offset': 0, 'w': WALL_DEPTH_CM, 'l': 0},
|
| 1010 |
+
'window': {'obj': 'wallWindow.obj', 'color': WINDOW_COLOR, 'height': 120, 'v_offset': 100, 'w': WALL_DEPTH_CM, 'l': 0}
|
| 1011 |
+
}
|
| 1012 |
+
|
| 1013 |
+
pool_items = {}
|
| 1014 |
+
if items_recomendados:
|
| 1015 |
+
for it in items_recomendados:
|
| 1016 |
+
pool_items.setdefault(it['Tipo_mueble'], []).append(it)
|
| 1017 |
+
|
| 1018 |
+
colores = {
|
| 1019 |
+
'Sofás': '#7f8c8d', 'Muebles de salón': '#95a5a6',
|
| 1020 |
+
'Mesas bajas de salón, de centro y auxiliares': '#d6bfa9',
|
| 1021 |
+
'Sillones': '#5d6d7e', 'Estanterías y librerías': '#ecf0f1'
|
| 1022 |
+
}
|
| 1023 |
+
|
| 1024 |
+
polygon_points = room_data.get('polygon_points', [])
|
| 1025 |
+
poly_pts_cm = np.array(polygon_points) * 100
|
| 1026 |
+
data = []
|
| 1027 |
+
|
| 1028 |
+
# --- FASE 1: DIBUJAR ESTRUCTURA DE LA HABITACIÓN (PAREDES Y OBSTÁCULOS) ---
|
| 1029 |
+
if len(poly_pts_cm) > 1:
|
| 1030 |
+
num_walls = len(poly_pts_cm)
|
| 1031 |
+
all_obstacles = room_data.get('doors', []) + room_data.get('windows', [])
|
| 1032 |
+
|
| 1033 |
+
for i in range(num_walls):
|
| 1034 |
+
p1 = poly_pts_cm[i]
|
| 1035 |
+
p2 = poly_pts_cm[(i+1) % num_walls]
|
| 1036 |
+
|
| 1037 |
+
length, cx_seg, cy_seg, angle = get_segment_properties(p1, p2)
|
| 1038 |
+
|
| 1039 |
+
# Identificar obstáculos en esta pared (índice i) y ordenarlos
|
| 1040 |
+
wall_obstacles = []
|
| 1041 |
+
for obs in all_obstacles:
|
| 1042 |
+
# Nota: obs['center'][1] es el índice normalizado de pared (0 a 1)
|
| 1043 |
+
wall_idx_obs = int(round(obs['center'][1] * num_walls)) % num_walls
|
| 1044 |
+
if wall_idx_obs == i:
|
| 1045 |
+
obs['type'] = 'door' if 'doors' in room_data and obs in room_data['doors'] else 'window'
|
| 1046 |
+
wall_obstacles.append(obs)
|
| 1047 |
+
wall_obstacles.sort(key=lambda x: x['center'][0]) # Ordenar por posición a lo largo de la pared
|
| 1048 |
+
|
| 1049 |
+
# Definir segmentos de pared a dibujar
|
| 1050 |
+
segments_to_draw = []
|
| 1051 |
+
current_start_pct = 0.0
|
| 1052 |
+
|
| 1053 |
+
for obs in wall_obstacles:
|
| 1054 |
+
center_pct = obs['center'][0]
|
| 1055 |
+
width_m = obs['width']
|
| 1056 |
+
width_pct = (width_m * 100) / length
|
| 1057 |
+
|
| 1058 |
+
obs_start_pct = max(0.0, center_pct - width_pct / 2)
|
| 1059 |
+
obs_end_pct = min(1.0, center_pct + width_pct / 2)
|
| 1060 |
+
|
| 1061 |
+
# Segmento de pared antes del obstáculo (pared vacía)
|
| 1062 |
+
if obs_start_pct > current_start_pct:
|
| 1063 |
+
segments_to_draw.append({'type': 'wall', 'start': current_start_pct, 'end': obs_start_pct})
|
| 1064 |
+
|
| 1065 |
+
# Segmento de obstáculo
|
| 1066 |
+
segments_to_draw.append({'type': obs['type'], 'start': obs_start_pct, 'end': obs_end_pct, 'w': width_m * 100})
|
| 1067 |
+
|
| 1068 |
+
current_start_pct = obs_end_pct
|
| 1069 |
+
|
| 1070 |
+
# Segmento de pared final (pared vacía)
|
| 1071 |
+
if current_start_pct < 1.0:
|
| 1072 |
+
segments_to_draw.append({'type': 'wall', 'start': current_start_pct, 'end': 1.0})
|
| 1073 |
+
|
| 1074 |
+
# --- DIBUJAR LOS SEGMENTOS CON OBJS ---
|
| 1075 |
+
|
| 1076 |
+
for seg in segments_to_draw:
|
| 1077 |
+
seg_start_cm = seg['start'] * length
|
| 1078 |
+
seg_end_cm = seg['end'] * length
|
| 1079 |
+
seg_len = seg_end_cm - seg_start_cm
|
| 1080 |
+
|
| 1081 |
+
if seg_len < 1: continue
|
| 1082 |
+
|
| 1083 |
+
# Recalcular centro y ángulo para el subsegmento
|
| 1084 |
+
seg_mid_x = p1[0] + (seg['start'] + seg['end']) / 2 * (p2[0] - p1[0])
|
| 1085 |
+
seg_mid_y = p1[1] + (seg['start'] + seg['end']) / 2 * (p2[1] - p1[1])
|
| 1086 |
+
|
| 1087 |
+
# Configuración del modelo
|
| 1088 |
+
if seg['type'] == 'wall':
|
| 1089 |
+
obj_file = 'wall.obj'
|
| 1090 |
+
color = WALL_COLOR
|
| 1091 |
+
h_val = WALL_H
|
| 1092 |
+
z_base = 0
|
| 1093 |
+
w_seg = seg_len # El largo del segmento es el ANCHO (X) del OBJ
|
| 1094 |
+
l_seg = WALL_DEPTH_CM # La profundidad de la pared es el LARGO (Y) del OBJ
|
| 1095 |
+
else:
|
| 1096 |
+
obs_data = OBSTACLE_MAP[seg['type']]
|
| 1097 |
+
obj_file = obs_data['obj']
|
| 1098 |
+
color = obs_data['color']
|
| 1099 |
+
h_val = obs_data['height']
|
| 1100 |
+
z_base = obs_data['v_offset']
|
| 1101 |
+
w_seg = seg_len
|
| 1102 |
+
l_seg = WALL_DEPTH_CM
|
| 1103 |
+
|
| 1104 |
+
# Cargar y transformar el OBJ
|
| 1105 |
+
|
| 1106 |
+
# Lista de elementos a dibujar en este segmento (puede ser múltiple para ventanas/puertas)
|
| 1107 |
+
sub_elements = []
|
| 1108 |
+
|
| 1109 |
+
if seg['type'] == 'wall':
|
| 1110 |
+
sub_elements.append({
|
| 1111 |
+
'obj': 'wall.obj', 'color': WALL_COLOR,
|
| 1112 |
+
'h': WALL_H, 'z': 0,
|
| 1113 |
+
'w': seg_len, 'l': WALL_DEPTH_CM,
|
| 1114 |
+
'name': 'Pared'
|
| 1115 |
+
})
|
| 1116 |
+
else:
|
| 1117 |
+
obs_data = OBSTACLE_MAP[seg['type']]
|
| 1118 |
+
|
| 1119 |
+
# 1. El Obstáculo en sí
|
| 1120 |
+
sub_elements.append({
|
| 1121 |
+
'obj': obs_data['obj'], 'color': obs_data['color'],
|
| 1122 |
+
'h': obs_data['height'], 'z': obs_data['v_offset'],
|
| 1123 |
+
'w': seg_len, 'l': WALL_DEPTH_CM,
|
| 1124 |
+
'name': seg['type'].title()
|
| 1125 |
+
})
|
| 1126 |
+
|
| 1127 |
+
# 2. Relleno SUPERIOR (Dintel) - Si hay espacio hasta el techo
|
| 1128 |
+
top_gap = WALL_H - (obs_data['v_offset'] + obs_data['height'])
|
| 1129 |
+
if top_gap > 1:
|
| 1130 |
+
sub_elements.append({
|
| 1131 |
+
'obj': 'wall.obj', 'color': WALL_COLOR,
|
| 1132 |
+
'h': top_gap, 'z': obs_data['v_offset'] + obs_data['height'],
|
| 1133 |
+
'w': seg_len, 'l': WALL_DEPTH_CM,
|
| 1134 |
+
'name': 'Muro Superior'
|
| 1135 |
+
})
|
| 1136 |
+
|
| 1137 |
+
# 3. Relleno INFERIOR (Antepecho) - Si el obstáculo no empieza en el suelo
|
| 1138 |
+
bottom_gap = obs_data['v_offset']
|
| 1139 |
+
if bottom_gap > 1:
|
| 1140 |
+
sub_elements.append({
|
| 1141 |
+
'obj': 'wall.obj', 'color': WALL_COLOR,
|
| 1142 |
+
'h': bottom_gap, 'z': 0,
|
| 1143 |
+
'w': seg_len, 'l': WALL_DEPTH_CM,
|
| 1144 |
+
'name': 'Muro Inferior'
|
| 1145 |
+
})
|
| 1146 |
+
|
| 1147 |
+
# Renderizar todos los sub-elementos del segmento
|
| 1148 |
+
for el in sub_elements:
|
| 1149 |
+
x_seg, y_seg, z_seg, i_seg, j_seg, k_seg = load_and_transform_mesh(
|
| 1150 |
+
el['obj'], w=el['w'], l=el['l'], h=el['h'],
|
| 1151 |
+
cx=seg_mid_x, cy=seg_mid_y, angle=angle, base_z=el['z']
|
| 1152 |
+
)
|
| 1153 |
+
|
| 1154 |
+
if x_seg is not None:
|
| 1155 |
+
data.append(go.Mesh3d(
|
| 1156 |
+
x=x_seg, y=y_seg, z=z_seg, i=i_seg, j=j_seg, k=k_seg,
|
| 1157 |
+
color=el['color'], opacity=1.0, flatshading=True,
|
| 1158 |
+
name=f'{el["name"]} P{i}', showlegend=(seg['type']!='wall' and el['name'] == seg['type'].title())
|
| 1159 |
+
))
|
| 1160 |
+
else:
|
| 1161 |
+
print(f"!!! FALLO DE CARGA: Omisión de {el['name']} en pared {i}.")
|
| 1162 |
+
|
| 1163 |
+
# 5. ESQUINA (wallCorner.obj) - Se dibuja en el vértice P1
|
| 1164 |
+
if i == 0 or True: # Dibujar esquina en cada vértice para cerrar bien
|
| 1165 |
+
# Nota: Dibujamos esquina en P1 (inicio del segmento)
|
| 1166 |
+
x_c, y_c, z_c, i_c, j_c, k_c = load_and_transform_mesh(
|
| 1167 |
+
'wallCorner.obj', w=WALL_DEPTH_CM, l=WALL_DEPTH_CM, h=WALL_H,
|
| 1168 |
+
cx=p1[0], cy=p1[1], angle=angle
|
| 1169 |
+
)
|
| 1170 |
+
|
| 1171 |
+
if x_c is not None:
|
| 1172 |
+
data.append(go.Mesh3d(
|
| 1173 |
+
x=x_c, y=y_c, z=z_c, i=i_c, j=i_c, k=i_c,
|
| 1174 |
+
color=CORNER_COLOR, opacity=1.0, flatshading=True,
|
| 1175 |
+
name=f'Esquina {i}', showlegend=False
|
| 1176 |
+
))
|
| 1177 |
+
|
| 1178 |
+
|
| 1179 |
+
# --- FASE 2: DIBUJAR SUELO ---
|
| 1180 |
+
# Usar un fill poly simple. Convertir a (X, Y, Z) para Plotly
|
| 1181 |
+
x_floor = poly_pts_cm[:, 0]
|
| 1182 |
+
y_floor = poly_pts_cm[:, 1]
|
| 1183 |
+
z_floor = np.zeros_like(x_floor)
|
| 1184 |
+
|
| 1185 |
+
# Crear caras del suelo (convexhull)
|
| 1186 |
+
from scipy.spatial import ConvexHull
|
| 1187 |
+
try:
|
| 1188 |
+
hull = ConvexHull(np.array([x_floor, y_floor]).T)
|
| 1189 |
+
i_f, j_f, k_f = hull.simplices.T
|
| 1190 |
+
except: # Fails if points are collinear or too few
|
| 1191 |
+
i_f, j_f, k_f = [], [], []
|
| 1192 |
+
|
| 1193 |
+
suelo_trace = go.Mesh3d(
|
| 1194 |
+
x=x_floor, y=y_floor, z=z_floor,
|
| 1195 |
+
i=i_f, j=j_f, k=k_f,
|
| 1196 |
+
color='#fafafa', opacity=1, name='Suelo', hoverinfo='skip'
|
| 1197 |
+
)
|
| 1198 |
+
data.append(suelo_trace)
|
| 1199 |
+
|
| 1200 |
+
# --- FASE 3: DIBUJAR MUEBLES (OBJ o OMISIÓN) ---
|
| 1201 |
+
for mueble in layout_plan:
|
| 1202 |
+
tipo = mueble['tipo']
|
| 1203 |
+
obj_file = MODEL_MAP.get(tipo)
|
| 1204 |
+
# Buscar la info real del mueble seleccionado
|
| 1205 |
+
info_real = pool_items.get(tipo, [{}])[0] if tipo in pool_items else {}
|
| 1206 |
+
|
| 1207 |
+
nombre_display = info_real.get('Nombre', tipo)
|
| 1208 |
+
precio = info_real.get('Precio', '?')
|
| 1209 |
+
desc = info_real.get('Descripcion', '')[:60]
|
| 1210 |
+
|
| 1211 |
+
hover_text = (
|
| 1212 |
+
f"<b>TIPO:</b> {tipo}<br>"
|
| 1213 |
+
f"<b>MODELO:</b> {nombre_display}<br>"
|
| 1214 |
+
f"<b>PRECIO:</b> {precio}€<br>"
|
| 1215 |
+
f"<i>{desc}...</i>"
|
| 1216 |
+
)
|
| 1217 |
+
|
| 1218 |
+
try:
|
| 1219 |
+
h_val = float(info_real.get('Altura', 60))
|
| 1220 |
+
if np.isnan(h_val) or h_val <= 0: h_val = 60
|
| 1221 |
+
except: h_val = 60
|
| 1222 |
+
|
| 1223 |
+
w_m = mueble['ancho']
|
| 1224 |
+
l_m = mueble['largo']
|
| 1225 |
+
|
| 1226 |
+
# Rotación extra para sofás (suelen venir mirando hacia atrás)
|
| 1227 |
+
rot_offset = np.pi if tipo == 'Sofás' else 0
|
| 1228 |
+
|
| 1229 |
+
# Lógica específica para detectar Sofás en L (Chaise Longue / Rinconera)
|
| 1230 |
+
if tipo == 'Sofás':
|
| 1231 |
+
keywords_l_shape = ['chaise', 'esquina', 'rincon', 'l-shaped', 'modular', 'l shape']
|
| 1232 |
+
text_to_search = (nombre_display + " " + desc).lower()
|
| 1233 |
+
if any(k in text_to_search for k in keywords_l_shape):
|
| 1234 |
+
obj_file = 'loungeDesignSofaCorner.obj'
|
| 1235 |
+
# Ajuste de rotación específico para este modelo si es necesario (a veces los modelos de esquina tienen otra orientación)
|
| 1236 |
+
# Por ahora mantenemos la rotación de sofá estándar (pi) o ajustamos si el usuario reporta algo raro.
|
| 1237 |
+
# rot_offset = np.pi
|
| 1238 |
+
|
| 1239 |
+
if obj_file:
|
| 1240 |
+
x_m, y_m, z_m, i_m, j_m, k_m = load_and_transform_mesh(
|
| 1241 |
+
obj_file, w=w_m, l=l_m, h=h_val,
|
| 1242 |
+
cx=mueble['x'], cy=mueble['y'], angle=mueble['angle'],
|
| 1243 |
+
rotation_offset=rot_offset
|
| 1244 |
+
)
|
| 1245 |
+
|
| 1246 |
+
if x_m is not None:
|
| 1247 |
+
traces = [go.Mesh3d(
|
| 1248 |
+
x=x_m, y=y_m, z=z_m, i=i_m, j=j_m, k=k_m,
|
| 1249 |
+
color=colores.get(tipo, '#95a5a6'), opacity=1.0, flatshading=True,
|
| 1250 |
+
name=nombre_display, hoverinfo='text', text=hover_text,
|
| 1251 |
+
lighting=dict(ambient=0.6, diffuse=0.8), showlegend=True
|
| 1252 |
+
)]
|
| 1253 |
+
data.extend(traces)
|
| 1254 |
+
else:
|
| 1255 |
+
# Fallo de carga: OMITIR
|
| 1256 |
+
print(f"!!! FALLO DE CARGA/RENDERIZADO: {tipo} ({nombre_display}). Modelo OBJ no usado.")
|
| 1257 |
+
continue
|
| 1258 |
+
else:
|
| 1259 |
+
# Omisión si no hay OBJ mapeado
|
| 1260 |
+
print(f"!!! OMISIÓN: No hay OBJ mapeado para el tipo: {tipo}. Saltando renderizado.")
|
| 1261 |
+
continue
|
| 1262 |
+
|
| 1263 |
+
# --- CONFIGURACIÓN FINAL ---
|
| 1264 |
+
max_dim = np.max(poly_pts_cm, axis=0) if poly_pts_cm.size > 0 else [500, 500]
|
| 1265 |
+
|
| 1266 |
+
layout = go.Layout(
|
| 1267 |
+
title="Diseño 3D (Interactúa con el ratón)",
|
| 1268 |
+
showlegend=True,
|
| 1269 |
+
scene=dict(
|
| 1270 |
+
xaxis=dict(visible=False), yaxis=dict(visible=False), zaxis=dict(visible=False),
|
| 1271 |
+
aspectmode='data', # Usar 'data' para que los ejes sean proporcionales a los valores reales
|
| 1272 |
+
aspectratio=None,
|
| 1273 |
+
bgcolor='white',
|
| 1274 |
+
camera=dict(eye=dict(x=2.0, y=2.0, z=2.0)) # Zoom out inicial
|
| 1275 |
+
),
|
| 1276 |
+
margin=dict(r=0, l=0, b=0, t=30),
|
| 1277 |
+
height=600
|
| 1278 |
+
)
|
| 1279 |
+
|
| 1280 |
+
return go.Figure(data=data, layout=layout)
|
| 1281 |
+
|
| 1282 |
+
|
| 1283 |
+
def generar_diagrama_planta(room_data):
|
| 1284 |
+
"""Genera un diagrama de planta 2D de la habitación con paredes, puertas y ventanas."""
|
| 1285 |
+
try:
|
| 1286 |
+
# --- 1. CONFIGURACIÓN DE ESTILO ---
|
| 1287 |
+
COLORS = {
|
| 1288 |
+
'bg': '#1C4E80', # Azul oscuro de fondo
|
| 1289 |
+
'line': '#ffffff', # Blanco para líneas
|
| 1290 |
+
'hole': '#1C4E80', # Mismo color que el fondo para "borrar" la pared
|
| 1291 |
+
'grid': '#ffffff',
|
| 1292 |
+
'text': '#ffffff'
|
| 1293 |
+
}
|
| 1294 |
+
|
| 1295 |
+
WALL_WIDTH = 6
|
| 1296 |
+
HOLE_WIDTH = 8
|
| 1297 |
+
ELEM_WIDTH = 1.5
|
| 1298 |
+
|
| 1299 |
+
fig, ax = plt.subplots(figsize=(12, 8))
|
| 1300 |
+
fig.patch.set_facecolor(COLORS['bg'])
|
| 1301 |
+
ax.set_facecolor(COLORS['bg'])
|
| 1302 |
+
|
| 1303 |
+
polygon_points = room_data.get('polygon_points', None)
|
| 1304 |
+
if polygon_points is None or len(polygon_points) < 3:
|
| 1305 |
+
ax.text(0.5, 0.5, "ERROR: Polígono de habitación inválido", color=COLORS['text'], ha='center')
|
| 1306 |
+
return fig
|
| 1307 |
+
|
| 1308 |
+
polygon_m = np.array(polygon_points)
|
| 1309 |
+
centroid = np.mean(polygon_m, axis=0)
|
| 1310 |
+
|
| 1311 |
+
# A. Forzar proporción real (1 metro visual = 1 metro dato)
|
| 1312 |
+
ax.set_aspect('equal', adjustable='box')
|
| 1313 |
+
|
| 1314 |
+
# B. Calcular límites enteros para asegurar que el grid cae en el metro exacto
|
| 1315 |
+
min_x, min_y = np.min(polygon_m, axis=0)
|
| 1316 |
+
max_x, max_y = np.max(polygon_m, axis=0)
|
| 1317 |
+
|
| 1318 |
+
# Márgenes de 1 metro extra alrededor
|
| 1319 |
+
start_x = np.floor(min_x - 1)
|
| 1320 |
+
end_x = np.ceil(max_x + 1)
|
| 1321 |
+
start_y = np.floor(min_y - 1)
|
| 1322 |
+
end_y = np.ceil(max_y + 1)
|
| 1323 |
+
|
| 1324 |
+
ax.set_xlim(start_x, end_x)
|
| 1325 |
+
ax.set_ylim(start_y, end_y)
|
| 1326 |
+
|
| 1327 |
+
# C. Definir ticks explícitamente cada 1.0 unidades (1 metro)
|
| 1328 |
+
xticks = np.arange(start_x, end_x + 1, 1.0)
|
| 1329 |
+
yticks = np.arange(start_y, end_y + 1, 1.0)
|
| 1330 |
+
|
| 1331 |
+
ax.set_xticks(xticks)
|
| 1332 |
+
ax.set_yticks(yticks)
|
| 1333 |
+
|
| 1334 |
+
# Grid muy sutil
|
| 1335 |
+
ax.grid(True, color=COLORS['grid'], linestyle=':', linewidth=0.5, alpha=0.2)
|
| 1336 |
+
ax.set_xticklabels([]); ax.set_yticklabels([])
|
| 1337 |
+
ax.tick_params(length=0)
|
| 1338 |
+
|
| 1339 |
+
# --- Helper: Datos de Muro (Necesario aquí para Doors/Windows) ---
|
| 1340 |
+
num_walls = len(polygon_m)
|
| 1341 |
+
def get_wall_data(idx, pct):
|
| 1342 |
+
idx = idx % num_walls
|
| 1343 |
+
p1, p2 = polygon_m[idx], polygon_m[(idx + 1) % num_walls]
|
| 1344 |
+
vec = p2 - p1
|
| 1345 |
+
L_wall = np.linalg.norm(vec)
|
| 1346 |
+
if L_wall == 0: return None
|
| 1347 |
+
unit = vec / L_wall
|
| 1348 |
+
center_on_wall = p1 + vec * (pct / 100.0)
|
| 1349 |
+
|
| 1350 |
+
# Vector normal hacia adentro
|
| 1351 |
+
n1 = np.array([-unit[1], unit[0]])
|
| 1352 |
+
# Comprobar si n1 apunta hacia el centroide
|
| 1353 |
+
if np.linalg.norm((center_on_wall + n1) - centroid) > np.linalg.norm((center_on_wall - n1) - centroid):
|
| 1354 |
+
normal_in = -n1
|
| 1355 |
+
else:
|
| 1356 |
+
normal_in = n1
|
| 1357 |
+
return center_on_wall, unit, normal_in
|
| 1358 |
+
|
| 1359 |
+
# --- 3. DIBUJAR PAREDES ---
|
| 1360 |
+
polygon_closed = np.vstack([polygon_m, polygon_m[0]])
|
| 1361 |
+
|
| 1362 |
+
# Relleno muy sutil del suelo
|
| 1363 |
+
ax.fill(polygon_closed[:, 0], polygon_closed[:, 1], color=COLORS['line'], alpha=0.05, zorder=1)
|
| 1364 |
+
|
| 1365 |
+
# EL MURO GRUESO
|
| 1366 |
+
ax.plot(polygon_closed[:, 0], polygon_closed[:, 1],
|
| 1367 |
+
color=COLORS['line'], linewidth=WALL_WIDTH, zorder=2, solid_capstyle='round')
|
| 1368 |
+
|
| 1369 |
+
|
| 1370 |
+
# --- 4. PUERTAS (Hueco + Hoja + Arco) ---
|
| 1371 |
+
for d in room_data.get('doors', []):
|
| 1372 |
+
# Obtener índice de pared y posición porcentual a lo largo de esa pared
|
| 1373 |
+
wall_idx = int(round(d['center'][1] * num_walls)) % num_walls
|
| 1374 |
+
pos_pct = d['center'][0] * 100 # a cm
|
| 1375 |
+
|
| 1376 |
+
info = get_wall_data(wall_idx, pos_pct)
|
| 1377 |
+
if not info: continue
|
| 1378 |
+
center, unit, normal_in = info
|
| 1379 |
+
w = d['width'] # ancho de la puerta en metros
|
| 1380 |
+
|
| 1381 |
+
# A. HUECO (Borrar muro)
|
| 1382 |
+
h_s = center - unit * (w/2)
|
| 1383 |
+
h_e = center + unit * (w/2)
|
| 1384 |
+
ax.plot([h_s[0], h_e[0]], [h_s[1], h_e[1]],
|
| 1385 |
+
color=COLORS['hole'], linewidth=HOLE_WIDTH, zorder=5)
|
| 1386 |
+
|
| 1387 |
+
# B. HOJA
|
| 1388 |
+
hinge = h_s
|
| 1389 |
+
tip = hinge + normal_in * w
|
| 1390 |
+
ax.plot([hinge[0], tip[0]], [hinge[1], tip[1]],
|
| 1391 |
+
color=COLORS['line'], linewidth=ELEM_WIDTH, zorder=6)
|
| 1392 |
+
|
| 1393 |
+
# C. ARCO (Interpolado)
|
| 1394 |
+
arc_pts = []
|
| 1395 |
+
start_angle = np.arctan2(unit[1], unit[0])
|
| 1396 |
+
|
| 1397 |
+
# Crear la rotación desde el vector 'unit' al vector 'normal_in'
|
| 1398 |
+
for t in np.linspace(0, 1, 15):
|
| 1399 |
+
angle_interp = t * (np.pi/2)
|
| 1400 |
+
# Aplicar rotación al vector 'unit'
|
| 1401 |
+
v_rot = unit * np.cos(angle_interp) + normal_in * np.sin(angle_interp)
|
| 1402 |
+
pt = hinge + v_rot * w
|
| 1403 |
+
arc_pts.append(pt)
|
| 1404 |
+
arc_pts = np.array(arc_pts)
|
| 1405 |
+
ax.plot(arc_pts[:, 0], arc_pts[:, 1],
|
| 1406 |
+
color=COLORS['line'], linestyle=':', linewidth=1, zorder=6)
|
| 1407 |
+
|
| 1408 |
+
# --- 5. VENTANAS (Hueco + Rectángulo vacío) ---
|
| 1409 |
+
for w_obj in room_data.get('windows', []):
|
| 1410 |
+
wall_idx = int(round(w_obj['center'][1] * num_walls)) % num_walls
|
| 1411 |
+
pos_pct = w_obj['center'][0] * 100
|
| 1412 |
+
|
| 1413 |
+
info = get_wall_data(wall_idx, pos_pct)
|
| 1414 |
+
if not info: continue
|
| 1415 |
+
center, unit, normal_in = info
|
| 1416 |
+
w = w_obj['width'] # ancho de la ventana en metros
|
| 1417 |
+
|
| 1418 |
+
# A. HUECO (Borrar muro grueso)
|
| 1419 |
+
h_s = center - unit * (w/2); h_e = center + unit * (w/2)
|
| 1420 |
+
ax.plot([h_s[0], h_e[0]], [h_s[1], h_e[1]],
|
| 1421 |
+
color=COLORS['hole'], linewidth=HOLE_WIDTH, zorder=5)
|
| 1422 |
+
|
| 1423 |
+
# B. MARCO RECTANGULAR (Sin relleno)
|
| 1424 |
+
frame_depth = 0.1 # 10 cm de profundidad de marco (en metros)
|
| 1425 |
+
|
| 1426 |
+
# 4 Esquinas del rectángulo
|
| 1427 |
+
c1 = h_s - normal_in * (frame_depth/2)
|
| 1428 |
+
c2 = h_e - normal_in * (frame_depth/2)
|
| 1429 |
+
c3 = h_e + normal_in * (frame_depth/2)
|
| 1430 |
+
c4 = h_s + normal_in * (frame_depth/2)
|
| 1431 |
+
|
| 1432 |
+
# Dibujar perímetro (c1->c2->c3->c4->c1)
|
| 1433 |
+
rect_x = [c1[0], c2[0], c3[0], c4[0], c1[0]]
|
| 1434 |
+
rect_y = [c1[1], c2[1], c3[1], c4[1], c1[1]]
|
| 1435 |
+
|
| 1436 |
+
ax.plot(rect_x, rect_y, color=COLORS['line'], linewidth=ELEM_WIDTH, zorder=6)
|
| 1437 |
+
|
| 1438 |
+
# --- 6. ETIQUETAS Y LEYENDA ---
|
| 1439 |
+
legend_items = []
|
| 1440 |
+
for i in range(num_walls):
|
| 1441 |
+
p1, p2 = polygon_m[i], polygon_m[(i + 1) % num_walls]
|
| 1442 |
+
mid = (p1 + p2) / 2
|
| 1443 |
+
|
| 1444 |
+
# Vector hacia afuera para etiqueta
|
| 1445 |
+
vec_out = mid - centroid
|
| 1446 |
+
vec_out = vec_out / np.linalg.norm(vec_out)
|
| 1447 |
+
text_pos = mid + vec_out * 0.5 # 50cm afuera para no tocar el muro grueso
|
| 1448 |
+
|
| 1449 |
+
L = np.linalg.norm(p2 - p1)
|
| 1450 |
+
legend_items.append(f"P{i+1}: {L:.2f} m")
|
| 1451 |
+
|
| 1452 |
+
ax.text(text_pos[0], text_pos[1], f"P{i+1}", color=COLORS['bg'], fontsize=8, fontweight='bold',
|
| 1453 |
+
ha='center', va='center', zorder=10,
|
| 1454 |
+
bbox=dict(boxstyle='circle,pad=0.2', fc='white', ec='none'))
|
| 1455 |
+
|
| 1456 |
+
plt.subplots_adjust(right=0.70)
|
| 1457 |
+
|
| 1458 |
+
info_text = "HABITACIÓN\n(Grid 1x1m)\n\n" + "\n".join(legend_items)
|
| 1459 |
+
|
| 1460 |
+
fig.text(0.72, 0.5, info_text, fontsize=10, color=COLORS['text'],
|
| 1461 |
+
fontfamily='monospace', va='center',
|
| 1462 |
+
bbox=dict(boxstyle='square,pad=1', fc=COLORS['hole'], ec=COLORS['line']))
|
| 1463 |
+
|
| 1464 |
+
return fig
|
| 1465 |
+
|
| 1466 |
+
except Exception as e:
|
| 1467 |
+
print(f"Error planta: {e}")
|
| 1468 |
+
import traceback
|
| 1469 |
+
traceback.print_exc()
|
| 1470 |
+
return plt.figure()
|
packages.txt
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
libgl1
|
requirements.txt
CHANGED
|
@@ -1,3 +1,15 @@
|
|
| 1 |
-
|
| 2 |
pandas
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
streamlit
|
| 2 |
pandas
|
| 3 |
+
numpy
|
| 4 |
+
Pillow
|
| 5 |
+
torch
|
| 6 |
+
torchvision
|
| 7 |
+
transformers
|
| 8 |
+
scikit-learn
|
| 9 |
+
matplotlib
|
| 10 |
+
shapely
|
| 11 |
+
tqdm
|
| 12 |
+
plotly
|
| 13 |
+
pywavefront
|
| 14 |
+
scipy
|
| 15 |
+
opencv-python-headless
|
vectores_cache.pkl
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:ead8295d791f946f9ddb61bcfdcbe1b1a447c1307dfffd1eefb1d80dda0f20b0
|
| 3 |
+
size 1090945
|