fix: cleanup

Signed-off-by: ivelin <ivelin.eth@gmail.com>

app.py

@@ -16,13 +16,16 @@ model.to(device)
 
 def process_refexp(image: Image, prompt: str):
 
-    print(f"(image, prompt): {image}, {prompt}")
+    # extract PIL image from Gradio Image component
+    pil_image = image.value
+
+    print(f"(image, prompt): {pil_image}, {prompt}")
 
     # trim prompt to 80 characters and normalize to lowercase
     prompt = prompt[:80].lower()
 
     # prepare encoder inputs
-    pixel_values = processor(image, return_tensors="pt").pixel_values
+    pixel_values = processor(pil_image, return_tensors="pt").pixel_values
 
     # prepare decoder inputs
     task_prompt = "<s_refexp><s_prompt>{user_input}</s_prompt><s_refexp>"
@@ -53,9 +56,9 @@ def process_refexp(image: Image, prompt: str):
     bbox = processor.token2json(sequence)
     print(f"predicted bounding box: {bbox}")
 
-    print(f"image object: {image}")
-    print(f"image size: {image.size}")
-    width, height = image.size
+    print(f"image object: {pil_image}")
+    print(f"image size: {pil_image.size}")
+    width, height = pil_image.size
     print(f"image width, height: {width, height}")
     print(f"prompt: {sample['prompt']}")
 
@@ -70,7 +73,7 @@ def process_refexp(image: Image, prompt: str):
     shape = [(xmin, ymin), (xmax, ymax)]
 
     # create rectangle image
-    img1 = ImageDraw.Draw(image)
+    img1 = ImageDraw.Draw(pil_image)
     img1.rectangle(shape, outline="green", width=5)
     return image, bbox