Create app.py

app.py

@@ -0,0 +1,89 @@
+
+import streamlit as st
+import torch
+from torchvision import transforms
+from PIL import Image
+from transformers import AutoTokenizer, AutoModel
+import torch.nn as nn
+
+# === Model Setup ===
+class MultimodalBanglaClassifier(nn.Module):
+    def __init__(self, text_model_name='sagorsarker/bangla-bert-base', num_classes=5):
+        super(MultimodalBanglaClassifier, self).__init__()
+        self.text_model = AutoModel.from_pretrained(text_model_name)
+        for param in self.text_model.encoder.layer[:6].parameters():
+            param.requires_grad = False
+
+        from torchvision.models import efficientnet_b3, EfficientNet_B3_Weights
+        self.image_model = efficientnet_b3(weights=EfficientNet_B3_Weights.IMAGENET1K_V1)
+        self.image_model.classifier = nn.Identity()
+
+        self.proj = nn.Linear(768 + 1536, 512)
+        self.transformer_fusion = nn.TransformerEncoder(
+            nn.TransformerEncoderLayer(d_model=512, nhead=4, batch_first=True),
+            num_layers=2
+        )
+        self.classifier = nn.Sequential(
+            nn.Linear(512, 256),
+            nn.ReLU(),
+            nn.Dropout(0.3),
+            nn.Linear(256, num_classes)
+        )
+
+    def forward(self, input_ids, attention_mask, image):
+        text_feat = self.text_model(input_ids=input_ids, attention_mask=attention_mask).last_hidden_state[:, 0, :]
+        image_feat = self.image_model(image)
+        fused = self.proj(torch.cat((text_feat, image_feat), dim=1)).unsqueeze(1)
+        fused = self.transformer_fusion(fused).squeeze(1)
+        return self.classifier(fused)
+
+@st.cache_resource
+def load_model():
+    model = MultimodalBanglaClassifier()
+    model.load_state_dict(torch.load("bangla_disaster_model.pth", map_location=torch.device('cpu')))
+    model.eval()
+    return model
+
+def predict(model, tokenizer, image, caption):
+    transform = transforms.Compose([
+        transforms.Resize((224, 224)),
+        transforms.ToTensor(),
+        transforms.Normalize(mean=[0.485, 0.456, 0.406],
+                             std=[0.229, 0.224, 0.225])
+    ])
+
+    image = transform(image).unsqueeze(0)
+
+    encoded = tokenizer(
+        caption,
+        padding='max_length',
+        truncation=True,
+        max_length=128,
+        return_tensors='pt'
+    )
+
+    with torch.no_grad():
+        output = model(
+            input_ids=encoded['input_ids'],
+            attention_mask=encoded['attention_mask'],
+            image=image
+        )
+        pred_class = output.argmax(dim=1).item()
+        classes = ['HYD', 'MET', 'FD', 'EQ', 'OTHD']
+        return classes[pred_class]
+
+st.title("🌪️ Bangla Disaster Classifier")
+
+uploaded_file = st.file_uploader("Upload an image", type=['jpg', 'png', 'jpeg'])
+caption = st.text_area("Enter Bangla caption", "")
+
+if uploaded_file and caption:
+    img = Image.open(uploaded_file).convert("RGB")
+    st.image(img, caption="Uploaded Image", use_column_width=True)
+
+    with st.spinner("Predicting..."):
+        tokenizer = AutoTokenizer.from_pretrained("sagorsarker/bangla-bert-base")
+        model = load_model()
+        prediction = predict(model, tokenizer, img, caption)
+
+    st.success(f"✅ Predicted Disaster Class: **{prediction}**")