new vqa example

app.py

@@ -9,7 +9,7 @@ login(token=os.environ["HUGGINGFACE_TOKEN"])
 demo_imgs = [
     ["images/chinchilla_web-1024x683.jpg", "images/shiba-inu-dog-in-the-snow.jpg", "images/900.jpeg", "images/dogs.jpeg"],
     ["images/hummus.jpg", "images/london-underground-sign.jpg", "images/4645808729_2dfc59b6a5_z.jpg", "images/5944609705_4664531909_z.jpg"],
-    ["images/COCO_train2014_000000572279.jpg", "images/COCO_train2014_000000194806.jpg", "images/istockphoto-622434332-1024x1024.jpg", "images/11887_pesto-pasta_Rita-1x1-1-501c953b29074ab193e2b5ad36e64648.jpg"],
+    ["images/latte.jpg", "images/COCO_train2014_000000194806.jpg", "images/istockphoto-622434332-1024x1024.jpg", "images/11887_pesto-pasta_Rita-1x1-1-501c953b29074ab193e2b5ad36e64648.jpg"],
     [
         "images/bcee7a-20190225-a-london-underground-sign.jpg",
         "images/istockphoto-622434332-1024x1024.jpg",
@@ -31,8 +31,8 @@ demo_texts = [
         "Output: a lavish reception room with black and white tiled floor."
     ],
     [
-        "Question: What is written on the plane? Answer: 'Cargo' is written on it in red.",
-        "Question: What is the man trying to catch? Answer: The man is catching a white kite that his friend is flying. The two men are on a beach.",
+        "Question: What latte art is presented in the image? Answer: A swan latte art is presented in the image.",
+        "Question: What is the man trying to catch? Answer: The man is catching a white kite that his friend is flying.",
         "Question: What does the sign say? Answer: Congress Ave",
         "Question: What is this dish? Answer: This is pesto pasta topped with cheese and basil.",
     ],