Add sidebar

app.py

@@ -22,35 +22,52 @@ def load_image_embeddings():
 k = 5
 img_dir = './images'
 
-st.title("MedCLIP 🩺")
-st.image("./assets/logo.png", width=100)
-st.markdown("""Search for medical images with natural language powered by a CLIP model [[Model Card]](https://huggingface.co/flax-community/medclip-roco) finetuned on the
+st.sidebar.header("MedCLIP")
+st.sidebar.image("./assets/logo.png", width=100)
+st.sidebar.empty()
+st.sidebar.markdown("""Search for medical images with natural language powered by a CLIP model [[Model Card]](https://huggingface.co/flax-community/medclip-roco) finetuned on the
  [Radiology Objects in COntext (ROCO) dataset](https://github.com/razorx89/roco-dataset).""")
-st.markdown("""Example queries:
-* `ultrasound scans`
-* `pathology`
-* `pancreatic carcinoma`
-* `PET scan`""")
+st.sidebar.markdown("""Example queries:
+* `ultrasound scans`🔍
+* `pathology`🔍
+* `pancreatic carcinoma`🔍
+* `PET scan`🔍""")
+k_slider = st.sidebar.slider("Number of images", min_value=1, max_value=10, value=5)
+st.sidebar.markdown("Kaushalya Madhawa, 2021")
+
+st.title("MedCLIP 🩺")
+# st.image("./assets/logo.png", width=100)
+# st.markdown("""Search for medical images with natural language powered by a CLIP model [[Model Card]](https://huggingface.co/flax-community/medclip-roco) finetuned on the
+#  [Radiology Objects in COntext (ROCO) dataset](https://github.com/razorx89/roco-dataset).""")
+# st.markdown("""Example queries:
+# * `ultrasound scans`🔍
+# * `pathology`🔍
+# * `pancreatic carcinoma`🔍
+# * `PET scan`🔍""")
 
 image_list, image_embeddings = load_image_embeddings()
 model, processor = load_model()
 
 query = st.text_input("Enter your query here:")
+dot_prod = None
+
+if st.button("Search") or k_slider:
+    if len(query)==0:
+        st.write("Please enter a valid search query")
+    else:
+        with st.spinner(f"Searching ROCO test set for {query}..."):
+            k = k_slider
+            inputs = processor(text=[query], images=None, return_tensors="jax", padding=True)
 
-if st.button("Search"):
-    with st.spinner(f"Searching ROCO test set for {query}..."):
-        inputs = processor(text=[query], images=None, return_tensors="jax", padding=True)
-
-        query_embedding = model.get_text_features(**inputs)
-        query_embedding = np.asarray(query_embedding)
-        query_embedding = query_embedding / np.linalg.norm(query_embedding, axis=-1, keepdims=True)
-        dot_prod = np.sum(np.multiply(query_embedding, image_embeddings), axis=1)
-        topk_images = dot_prod.argsort()[-k:]
-        matching_images = image_list[topk_images]
-        top_scores = 1. - dot_prod[topk_images]
-        #show images
-        for img_path, score in zip(matching_images, top_scores):
-            img = plt.imread(os.path.join(img_dir, img_path))
-            st.image(img, width=300)
-            st.write(f"{img_path} ({score:.2f})", help="score")
-        
+            query_embedding = model.get_text_features(**inputs)
+            query_embedding = np.asarray(query_embedding)
+            query_embedding = query_embedding / np.linalg.norm(query_embedding, axis=-1, keepdims=True)
+            dot_prod = np.sum(np.multiply(query_embedding, image_embeddings), axis=1)
+            topk_images = dot_prod.argsort()[-k:]
+            matching_images = image_list[topk_images]
+            top_scores = 1. - dot_prod[topk_images]
+            #show images
+            for img_path, score in zip(matching_images, top_scores):
+                img = plt.imread(os.path.join(img_dir, img_path))
+                st.image(img, width=300)
+                st.write(f"{img_path} ({score:.2f})", help="score")