Update app.py

app.py

@@ -380,7 +380,7 @@ with st.sidebar:
 
 
 st.markdown("<div class='header-title'>Skin Cancer Image Classifier</div>", unsafe_allow_html=True)
-st.markdown("<div class='header-sub'>CNN Classifier • Model Attention (Grad-CAM++) visualizations • VLM explanations</div>", unsafe_allow_html=True)
+st.markdown("<div class='header-sub'>CNN inference • Model Attention (Grad-CAM++) visualizations • VLM explanations</div>", unsafe_allow_html=True)
 
 uploaded_file = st.file_uploader("Upload a skin lesion image", type=["jpg","jpeg","png"], key="uploaded_file" )
 
@@ -425,40 +425,8 @@ if 'selected_image' in st.session_state:
         st.session_state["conf"] = conf
         st.session_state["overlay_pil"] = overlay_pil
         st.session_state["last_image_bytes"] = img_bytes
-        try:
-            with st.spinner("Loading VLM Model. Please be patient..."):
-                try:
-                    vlm_info = load_vlm_model()
-                except Exception as e:
-                    st.error("VLM load failed. See logs above.")
-                    vlm_info = None
-
-            if vlm_info is not None:
-                try:
-                    img_for_vlm = overlay_pil.convert("RGB").resize((224, 224), Image.BILINEAR)
-                except Exception:
-                    st.warning("Overlay image not available for VLM input; using original image.")
-                    img_for_vlm = pil_img.convert("RGB").resize((224, 224), Image.BILINEAR)
-
-                with st.spinner("Generating Explanation...."):
-                    response = generate_vlm_response(
-                        vlm_info["processor"],
-                        vlm_info["model"],
-                        vlm_info["device"],
-                        img_for_vlm,
-                        pred_label,
-                        max_new_tokens=128
-                    )
-                #response = "Debugging VLM response." # For debugging
-                if response is None:
-                    st.error("VLM did not return a response.")
-                else:
-                    st.session_state["vlm_response"] = response
-        except Exception as e:
-            st.error(f"Error in VLM generation flow: {e}")
-            st.exception(traceback.format_exc())
-
-
+        st.session_state["vlm_response"] = None
+        
 with attention_column:
     st.markdown("<div class='card'>", unsafe_allow_html=True)
     st.subheader("Model Attention Visualization")
@@ -478,12 +446,31 @@ else:
     c1.metric("Predicted", "—")
     c2.metric("Confidence", "—")
 
-# VLM Response placeholder
-st.subheader("Generated Explanation")
-if st.session_state.get("vlm_response"):
-    st.info(st.session_state["vlm_response"])
-else:
-    st.info("VLM explanation will appear here after selecting an image and running classification.")
+if st.button("Generate VLM Explanation"):
+    if 'selected_image' in st.session_state:
+        if not st.session_state.get("vlm_response", False):
+            try:
+                with st.spinner("Loading VLM Model. First time load will take time. Please be patient..."):
+                    
+                    response = "Debugging VLM response."
+                    if response is None:
+                        st.error("VLM did not return a response.")
+                    else:
+                        st.session_state["vlm_response"] = response
+
+            except Exception as e:
+                st.error(f"Error in VLM generation flow: {e}")
+                st.exception(traceback.format_exc())
+    else:
+        st.warning("Upload an image first or use the example images provided below!")
+
+
+if st.session_state.get("vlm_response", False):
+    st.subheader("Generated Explanation")
+    if st.session_state.get("vlm_response"):
+        st.info(st.session_state["vlm_response"])
+    else:
+        st.info("VLM explanation will appear here after selecting an image and running classification.")
 
 example_paths = [
     "images/ISIC_0025314.jpg",
@@ -563,7 +550,7 @@ if toggle:
                     st.toast(f"✅ Selected image: {selected_path}", icon="📸")
                 except Exception:
                     st.success(f"Selected image: {selected_path}")
-            st.rerun()
+                st.rerun()
 st.markdown("</div>", unsafe_allow_html=True)
 
 st.markdown("""