app.py

import gradio as gr
import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
from matplotlib.figure import Figure

# Charger ton modèle
model = tf.keras.models.load_model("MobileNetV2_UL_ML_c3_l0_acc97_auc100_20251012_161415.keras")

# Classes
classes = ["Cubisme", "Expressionnisme", "Post-impressionnisme"]

# Fonction de prédiction avec graphique personnalisé
"""def predire(image):
    # Prédiction
    image_resized = tf.image.resize(image, (224, 224)) / 255.0
    preds = model.predict(tf.expand_dims(image_resized, axis=0))[0]
    
    # Créer le graphique avec matplotlib
    fig = Figure(figsize=(10, 6))
    ax = fig.add_subplot(111)
    
    # Trier par probabilité décroissante
    sorted_indices = np.argsort(preds)[::-1]
    sorted_classes = [classes[i] for i in sorted_indices]
    sorted_probs = [preds[i] for i in sorted_indices]
    
    # Définir les couleurs : vert si > 50%, bleu sinon
    colors = ['#2ecc71' if prob >= 0.5 else '#bdc3c7' for prob in sorted_probs]
    
    # Créer le bar plot horizontal
    bars = ax.barh(sorted_classes, sorted_probs, color=colors, edgecolor='black', linewidth=1.5)
    
    # Ajouter les pourcentages sur les barres
    for i, (bar, prob) in enumerate(zip(bars, sorted_probs)):
        width = bar.get_width()
        label_x = width + 0.02 if width < 0.9 else width - 0.02
        ha = 'left' if width < 0.9 else 'right'
        text_color = 'black' if width < 0.9 else 'white'
        ax.text(label_x, bar.get_y() + bar.get_height()/2, 
                f'{prob*100:.1f}%', 
                ha=ha, va='center', fontsize=12, fontweight='bold', color=text_color)
    
    # Configuration du graphique
    ax.set_xlabel('Probabilité', fontsize=12, fontweight='bold')
    ax.set_xlim(0, 1.0)
    ax.set_title('Probabilités par mouvement pictural', fontsize=14, fontweight='bold', pad=20)
    ax.grid(axis='x', alpha=0.3, linestyle='--')
    ax.set_axisbelow(True)
    
    # Légende
    from matplotlib.patches import Patch
    legend_elements = [
        Patch(facecolor='#2ecc71', edgecolor='black', label='≥ 50%'),
        Patch(facecolor='#bdc3c7', edgecolor='black', label='< 50%')
    ]
    ax.legend(handles=legend_elements, loc='upper right', fontsize=10)
    
    fig.tight_layout()
    
    return fig"""

# Solution 2
"""def predire(image):
    image_resized = tf.image.resize(image, (224, 224)) / 255.0
    preds = model.predict(tf.expand_dims(image_resized, axis=0))[0]

    sorted_indices = np.argsort(preds)[::-1]
    sorted_classes = [classes[i] for i in sorted_indices]
    sorted_probs = [preds[i] for i in sorted_indices]
    colors = ['#2ecc71' if prob >= 0.5 else '#bdc3c7' for prob in sorted_probs]

    fig = Figure(figsize=(4, 3))  # Format compact adapté mobile
    ax = fig.add_subplot(111)

    # Barres verticales
    bars = ax.bar(sorted_classes, sorted_probs, color=colors, edgecolor='black', linewidth=1.5)

    # Ajout pourcentages
    for bar, prob in zip(bars, sorted_probs):
        ax.text(bar.get_x() + bar.get_width()/2, bar.get_height() + 0.03, 
                f"{prob*100:.1f}%", ha='center', va='bottom', fontsize=13, fontweight='bold', color='black')

    ax.set_ylabel('Probabilité', fontsize=15, fontweight='bold')
    ax.set_ylim(0, 1.0)
    ax.set_title('Probabilités par mouvement pictural', fontsize=16, fontweight='bold', pad=20)
    ax.grid(axis='y', alpha=0.15, linestyle='--')
    ax.set_axisbelow(True)

    # Titres inclinés à 45°
    ax.set_xticklabels(sorted_classes, rotation=45, ha='right', fontsize=15, fontweight='bold')
    fig.tight_layout()

    return fig"""

# Solution 3 : Passer par Plotly
"""import plotly.graph_objects as go

def predire(image):
    image_resized = tf.image.resize(image, (224, 224)) / 255.0
    preds = model.predict(tf.expand_dims(image_resized, axis=0))[0]

    sorted_indices = np.argsort(preds)[::-1]
    sorted_classes = [classes[i] for i in sorted_indices]
    sorted_probs = [preds[i] for i in sorted_indices]
    colors = ['#2ecc71' if prob >= 0.5 else '#bdc3c7' for prob in sorted_probs]

    fig = go.Figure(go.Bar(
        x=sorted_classes,
        y=sorted_probs,
        text=[f"{p*100:.1f}%" for p in sorted_probs],
        marker=dict(color=colors, line=dict(color='black', width=1)),
        textposition='auto',
    ))
    fig.update_layout(
        xaxis=dict(tickangle=45, tickfont=dict(size=17)),
        yaxis=dict(range=[0,1], title='Probabilité', tickfont=dict(size=17)),
        title="Probabilités par mouvement pictural",
        margin=dict(l=15, r=15, t=40, b=25),
        height=280,
        font=dict(size=17)
    )
    return fig"""

# Solution avec le graphique plus haut (ADAPTER LA HAUTEUR)
"""def predire(image):
    image_resized = tf.image.resize(image, (224, 224)) / 255.0
    preds = model.predict(tf.expand_dims(image_resized, axis=0))[0]
    sorted_indices = np.argsort(preds)[::-1]
    sorted_classes = [classes[i] for i in sorted_indices]
    sorted_probs = [preds[i] for i in sorted_indices]
    colors = ['#2ecc71' if prob >= 0.5 else '#bdc3c7' for prob in sorted_probs]

    fig = Figure(figsize=(12, 4.2))  # hauteur augmentée
    ax = fig.add_subplot(111)
    bars = ax.bar(sorted_classes, sorted_probs, color=colors, edgecolor='black', linewidth=1.5)

    for bar, prob in zip(bars, sorted_probs):
        ax.text(bar.get_x() + bar.get_width()/2, bar.get_height() + 0.035, 
                f"{prob*100:.1f}%", ha='center', va='bottom', fontsize=12, color='black', fontweight='bold')

    ax.set_ylabel('Probabilité', fontsize=14, fontweight='bold')
    ax.set_ylim(0, 1)
    ax.set_title("Probabilités par mouvement pictural", fontsize=14, fontweight='bold', pad=20)
    ax.set_xticklabels(sorted_classes, rotation=45, ha='right', fontsize=13, fontweight='bold')
    fig.tight_layout(pad=2.0)
    return fig"""

# Façon 5 uniquement sur le texte mais ça donne une erreur
"""def predire(image):
    image_resized = tf.image.resize(image, (224, 224)) / 255.0
    preds = model.predict(tf.expand_dims(image_resized, axis=0))[0]
    sorted_indices = np.argsort(preds)[::-1]
    sorted_classes = [classes[i] for i in sorted_indices]
    sorted_probs = [preds[i] for i in sorted_indices]

    # HTML pour Gradio : barres remplies & label
    bars = []
    for cls, prob in zip(sorted_classes, sorted_probs):
        color = "#2ecc71" if prob >= 0.5 else "#bdc3c7"
        bars.append(f'''
        <div style="background:linear-gradient(90deg,{color} {prob*100:.1f}%,#fff {prob*100:.1f}%);padding:6px 0;margin:5px 0;border-radius:4px;">
            <span style="padding-left:12px;font-size:15px;font-weight:bold;">{cls} — {prob*100:.1f} %</span>
        </div>''')
    return gr.HTML("".join(bars))

demo = gr.Interface(
    fn=predire,
    inputs=gr.Image(type="numpy", label="Importer une œuvre"),
    outputs=[gr.Image(label="Input"), gr.HTML(label="Résultats")],
    title="🎨 Classification de style pictural",
    examples=None,
    theme=gr.themes.Soft()
)"""

"""import plotly.graph_objects as go

def predire(image):
    image_resized = tf.image.resize(image, (224, 224)) / 255.0
    preds = model.predict(tf.expand_dims(image_resized, axis=0))[0]
    sorted_indices = np.argsort(preds)[::-1]
    sorted_classes = [classes[i] for i in sorted_indices]
    sorted_probs = [preds[i] for i in sorted_indices]
    colors = ['#2ecc71' if p >= 0.5 else '#bdc3c7' for p in sorted_probs]

    fig = go.Figure(go.Bar(
        x=sorted_classes,
        y=sorted_probs,
        marker=dict(color=colors, line=dict(color='black', width=1)),
        text=[f"{p*100:.1f}%" for p in sorted_probs],
        textposition='auto'
    ))
    fig.update_layout(
        xaxis=dict(fixedrange=True, tickangle=45, tickfont=dict(size=15)),
        yaxis=dict(fixedrange=True, range=[0,1], title="Probabilité", tickfont=dict(size=14)),
        title=dict(
            text="Probabilités par mouvement pictural",
            y=0.95,  # remonte le titre pour laisser de l’espace
            pad=dict(b=50)  # espace (en px) sous le titre, ajuste à volonté
        ),
        margin=dict(l=20, r=20, t=32, b=46),
        height=500,
        font=dict(size=16)
    )


    # Pour placer le texte à l’intérieur des barres
    fig.show(config={'displayModeBar': False})
    return fig
"""

import plotly.graph_objects as go

def predire(image):
    image_resized = tf.image.resize(image, (224, 224)) / 255.0
    preds = model.predict(tf.expand_dims(image_resized, axis=0))[0]
    sorted_indices = np.argsort(preds)[::-1]
    sorted_classes = [classes[i] for i in sorted_indices]
    sorted_probs = [preds[i] for i in sorted_indices]
    colors = ['#2ecc71' if p >= 0.5 else '#bdc3c7' for p in sorted_probs]

    fig = go.Figure(go.Bar(
        x=sorted_classes,
        y=sorted_probs,
        marker=dict(color=colors, line=dict(color='black', width=1)),
        text=[f"{p*100:.1f}%" for p in sorted_probs],
        textposition='auto'
    ))

    fig.update_layout(
        xaxis=dict(fixedrange=True, tickangle=45, tickfont=dict(size=15), automargin=True),
        yaxis=dict(fixedrange=True, range=[0, 1], title="Probabilité", tickfont=dict(size=14)),
        title=dict(
            text="Probabilités par <br>mouvement pictural",
            y=0.90,
            pad=dict(b=30)
        ),
        margin=dict(l=20, r=20, t=0, b=60),  # marge top plus haute et bottom plus grande
        height=600,
        font=dict(size=13)
    )

    # Pour placer le texte à l’intérieur des barres
    fig.data[0].textfont = dict(color='black', size=14, family="Arial")
    fig.show(config={'displayModeBar': False})
    return fig

# Interface Gradio
# Exemple pour Plotly hors Gradio :

demo = gr.Interface(
    fn=predire,
    inputs=gr.Image(type="numpy", label="Importer une œuvre"),
    outputs=gr.Plot(label="Résultats de la classification"),
    title="🎨 Classification de style pictural",
    description="Upload une image et découvre le mouvement pictural estimé par le CNN. Les barres vertes indiquent une probabilité supérieure ou égale à 50%.",
    examples=None,
    theme=gr.themes.Soft()
)

demo.launch()