Update app.py

app.py

@@ -1,15 +1,22 @@
-# app.py
-
 import re
 import gradio as gr
 from PIL import Image
-from transformers import pipeline
+from transformers import (
+    AutoProcessor,
+    AutoModelForVision2Seq,
+    pipeline,
+)
+
+# Load processor & model
+processor = AutoProcessor.from_pretrained("Salesforce/blip2-flan-t5-xl")
+model     = AutoModelForVision2Seq.from_pretrained("Salesforce/blip2-flan-t5-xl")
 
-# Single multi-modal pipeline: BLIP2 + Flan-T5-XL
+# Create the image-text-to-text pipeline
 pipe = pipeline(
     task="image-text-to-text",
-    model="Salesforce/blip2-flan-t5-xl",
-    tokenizer="Salesforce/blip2-flan-t5-xl",
+    model=model,
+    feature_extractor=processor.feature_extractor,
+    tokenizer=processor.tokenizer,
     max_new_tokens=500,
     do_sample=True,
     temperature=1.0,
@@ -17,7 +24,6 @@ pipe = pipeline(
     top_p=0.9,
 )
 
-# Hard-coded gallery URLs
 def get_recommendations():
     return [
         "https://i.imgur.com/InC88PP.jpeg",
@@ -45,27 +51,25 @@ def process(image: Image):
         "- <bullet 5>\n"
     )
 
-    # run the multi-modal pipeline
-    result = pipe(image, prompt=prompt)[0]["generated_text"]
+    out = pipe(image, prompt=prompt)[0]["generated_text"]
 
-    # extract the three parts via regex
-    cat = re.search(r"Category:(.*?)Analysis:", result, re.S)
-    ana = re.search(r"Analysis:(.*?)Suggestions:", result, re.S)
-    sug = re.search(r"Suggestions:(.*)", result, re.S)
+    # pull out each section via regex
+    cat = re.search(r"Category:(.*?)Analysis:", out, re.S)
+    ana = re.search(r"Analysis:(.*?)Suggestions:", out, re.S)
+    sug = re.search(r"Suggestions:(.*)", out, re.S)
 
     category    = cat.group(1).strip() if cat else ""
     analysis    = ana.group(1).strip() if ana else ""
     suggestions = sug.group(1).strip() if sug else ""
 
-    # ensure exactly five bullets
-    bullets = [line for line in suggestions.splitlines() if line.startswith("-")]
+    # ensure exactly 5 bullets
+    bullets = [l for l in suggestions.splitlines() if l.startswith("-")]
     if len(bullets) < 5:
         bullets += ["- (no suggestion)"] * (5 - len(bullets))
     suggestions = "\n".join(bullets[:5])
 
     return category, analysis, suggestions, get_recommendations()
 
-# build the Gradio interface
 with gr.Blocks(theme=gr.themes.Default(primary_hue="blue")) as demo:
     gr.Markdown("## 📢 Smart Ad Analyzer")
     gr.Markdown(
@@ -74,7 +78,7 @@ with gr.Blocks(theme=gr.themes.Default(primary_hue="blue")) as demo:
     )
 
     with gr.Row():
-        image_input = gr.Image(type="pil", label="Upload Ad Image")
+        img = gr.Image(type="pil", label="Upload Ad Image")
         with gr.Column():
             cat_out = gr.Textbox(label="Ad Category", interactive=False)
             ana_out = gr.Textbox(label="Ad Analysis", lines=5, interactive=False)
@@ -82,12 +86,7 @@ with gr.Blocks(theme=gr.themes.Default(primary_hue="blue")) as demo:
             btn     = gr.Button("Analyze Ad", size="sm", variant="primary")
 
     gallery = gr.Gallery(label="Recommended Example Ads", show_label=True)
-
-    btn.click(
-        fn=process,
-        inputs=[image_input],
-        outputs=[cat_out, ana_out, sug_out, gallery]
-    )
+    btn.click(fn=process, inputs=[img], outputs=[cat_out, ana_out, sug_out, gallery])
 
     gr.Markdown("Made by Simon Thalmay")