Update app.py

app.py

@@ -1,17 +1,15 @@
-# app.py
-
 import re
 import gradio as gr
 from PIL import Image
 from transformers import pipeline
 
-# Use ChatDOC/OCRFlux-3B for image-to-text instead of BLIP
+# 1) Image-to-text: ChatDOC/OCRFlux-3B for rich description
 image_to_text = pipeline(
     "image-to-text",
     model="ChatDOC/OCRFlux-3B"
 )
 
-# Helper to create Flan-T5 pipelines (temperature=1.0 for diversity)
+# 2) Helper to build Flan-T5-small text pipelines (temp=1.0)
 def make_pipeline(model_name, max_tokens):
     return pipeline(
         "text2text-generation",
@@ -19,15 +17,18 @@ def make_pipeline(model_name, max_tokens):
         tokenizer=model_name,
         max_new_tokens=max_tokens,
         do_sample=True,
-        temperature=1.0
+        temperature=1.0,
+        top_k=50,
+        top_p=0.95
     )
 
-# Pipelines: category, analysis, suggestions
+# 3) Pipelines: category, analysis, suggestions
 category_generator   = make_pipeline("google/flan-t5-small", 100)
 analysis_generator   = make_pipeline("google/flan-t5-small", 500)
 suggestion_generator = make_pipeline("google/flan-t5-small", 500)
 
-# Example ads URLs for gallery
+# Hardcoded example ads for gallery
+
 def get_recommendations():
     return [
         "https://i.imgur.com/InC88PP.jpeg",
@@ -42,18 +43,22 @@ def get_recommendations():
         "https://i.imgur.com/Xj92Cjv.jpeg",
     ]
 
-# Step 1: Use OCRFlux to get a detailed textual description of the image
+# Step A: Use OCRFlux to generate a detailed caption
+
 def generate_caption(image):
     result = image_to_text(image)
-    return result[0]["generated_text"].strip()
+    text = result[0]["generated_text"].strip()
+    return text
+
+# Step B: Flan interprets caption into concise category
 
-# Step 2: Flan interprets caption into a concise category label
 def generate_category(caption):
     prompt = f"Caption: {caption}\nProvide a concise category label for this ad."
     raw = category_generator(prompt)[0]["generated_text"].strip()
     return raw.splitlines()[0]
 
-# Step 3: Flan writes exactly five sentences of analysis
+# Step C: Flan produces exactly five-sentence analysis
+
 def generate_analysis(caption):
     prompt = (
         f"Caption: {caption}\n"
@@ -63,7 +68,8 @@ def generate_analysis(caption):
     sentences = re.split(r'(?<=[.!?])\s+', raw)
     return " ".join(sentences[:5])
 
-# Step 4: Flan suggests five bullet-point improvements
+# Step D: Flan suggests five actionable bullet-point improvements
+
 def generate_suggestions(caption):
     prompt = (
         f"Caption: {caption}\n"
@@ -74,13 +80,12 @@ def generate_suggestions(caption):
     lines = [l for l in raw.splitlines() if l.strip().startswith('- ')]
     if len(lines) < 5:
         all_lines = [l.strip() for l in raw.splitlines() if l.strip()]
-        lines = [
-            ('- ' + all_lines[i]) if not all_lines[i].startswith('- ') else all_lines[i]
-            for i in range(min(5, len(all_lines)))
-        ]
+        lines = [('- ' + all_lines[i]) if not all_lines[i].startswith('- ') else all_lines[i]
+                 for i in range(min(5, len(all_lines)))]
     return "\n".join(lines[:5])
 
-# Full workflow
+# Orchestrator: process image through all steps
+
 def process(image):
     caption     = generate_caption(image)
     category    = generate_category(caption)
@@ -89,11 +94,11 @@ def process(image):
     recs        = get_recommendations()
     return category, analysis, suggestions, recs
 
-# Gradio UI
+# Gradio UI layout
 with gr.Blocks(theme=gr.themes.Default(primary_hue="blue")) as demo:
     gr.Markdown("## 📢 Smart Ad Analyzer")
     gr.Markdown(
-        "Upload an image ad to see an Ad Category, a five-sentence Analysis, "
+        "Upload an image ad to see: an Ad Category, a five-sentence Analysis, "
         "five bullet-point Suggestions, and Example Ads."
     )
 
@@ -116,4 +121,4 @@ with gr.Blocks(theme=gr.themes.Default(primary_hue="blue")) as demo:
     gr.Markdown("Made by Simon Thalmay")
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()