feature: extract feature from user input image

image2text.py

@@ -1,5 +1,6 @@
 import streamlit as st
 import numpy as np
+import jax.numpy as jnp
 from PIL import Image
 
 from utils import load_model
@@ -22,8 +23,11 @@ def app(model_name):
             st.error("Please upload an image query.")
         else:
             image = Image.open(query)
-            inputs = processor(text=[""], images=image, return_tensors="jax", padding=True)
-            # vec = np.asarray(model.get_image_features(**inputs))
+            pixel_values = processor(
+                text=[""], images=image, return_tensors="jax", padding=True
+            ).pixel_values
+            pixel_values = jnp.transpose(pixel_values, axes=[0, 2, 3, 1])
+            vec = np.asarray(model.get_image_features(pixel_values))
             # ids, dists = index.knnQuery(vec, k=10)
             # result_files = map(lambda id: files[id], ids)
             # result_imgs, result_captions = [], []