app.py

import torch
from torchvision import transforms
from PIL import Image
import streamlit as st
import json
from torchvision.models import efficientnet_b7, EfficientNet_B7_Weights
import torch.nn as nn



# Charger les noms des classes
with open("class_names.json", "r") as f:
    class_names = json.load(f)
    
# Charger le modèle avec des poids pré-entraînés
weights = EfficientNet_B7_Weights.DEFAULT
base_model = efficientnet_b7(weights=weights)

# Adapter le modèle pour la classification
class CustomEfficientNet(nn.Module):
    def __init__(self, base_model, num_classes):
        super(CustomEfficientNet, self).__init__()
        self.base = nn.Sequential(*list(base_model.children())[:-2])
        self.global_avg_pool = nn.AdaptiveAvgPool2d(1)
        self.fc1 = nn.Linear(2560, 512)
        self.relu = nn.ReLU()
        self.fc2 = nn.Linear(512, num_classes)

    def forward(self, x):
        x = self.base(x)
        x = self.global_avg_pool(x)
        x = x.view(x.size(0), -1)
        x = self.relu(self.fc1(x))
        x = self.fc2(x)
        return x

# Définir le modèle final
num_classes = 2
model = CustomEfficientNet(base_model, num_classes).to("cuda" if torch.cuda.is_available() else "cpu")
model.load_state_dict(torch.load("efficientnet_b7_best.pth",weights_only=False))
model.eval()  # Passer le modèle en mode évaluation


# Définir la taille de l'image
image_size = (224, 224)

# Transformation pour l'image
class GrayscaleToRGB:
    def __call__(self, img):
        return img.convert("RGB")

valid_test_transforms = transforms.Compose([
    transforms.Grayscale(num_output_channels=1),
    transforms.Resize(image_size),
    GrayscaleToRGB(),  # Conversion en RGB
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])
])

# Fonction de prédiction
def predict_image(image):
    image_tensor = valid_test_transforms(image).unsqueeze(0).to(device)
    with torch.no_grad():
        outputs = model(image_tensor)
        _, predicted_class = torch.max(outputs, 1)
    predicted_label = class_names[predicted_class.item()]
    return predicted_label

# Interface Streamlit
st.title("Prédiction d'images avec PyTorch")
st.write("Chargez une image pour obtenir une prédiction de classe.")

uploaded_image = st.file_uploader("Téléchargez une image", type=["jpg", "jpeg", "png"])

if uploaded_image is not None:
    image = Image.open(uploaded_image)
    st.image(image, caption="Image téléchargée", use_column_width=True)
    
    predicted_label = predict_image(image)
    st.write(f"Prédiction de la classe : {predicted_label}")