commit20

app.py

@@ -83,6 +83,13 @@ from PIL import Image, ImageOps
 import numpy as np
 from PIL import Image, ImageOps
 import numpy as np
+from PIL import Image, ImageOps
+import numpy as np
+
+from transformers import CLIPProcessor, CLIPModel
+
+clip_model = CLIPModel.from_pretrained("openai/clip-vit-base-patch32").to(device)
+clip_processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch32")
 
 def extract_metadata_filters(query: str):
     query_lower = query.lower()
@@ -97,7 +104,8 @@ def extract_metadata_filters(query: str):
         "women": "Women", "woman": "Women", "womens": "Women", "female": "Women",
         "boys": "Boys", "boy": "Boys",
         "girls": "Girls", "girl": "Girls",
-        "kids": "Kids", "unisex": "Unisex"
+        "kids": "Kids","kid": "Kids",
+        "unisex": "Unisex"
     }
     for term, mapped_value in gender_map.items():
         if term in query_lower:
@@ -115,7 +123,7 @@ def extract_metadata_filters(query: str):
         "trousers": "Trousers", "pants": "Trousers",
         "shorts": "Shorts",
         "footwear": "Footwear",
-        "shoes": "Footwear",
+        "shoes": "Shoes",   # note kept as Shoes
         "fashion": "Apparel"
     }
     for term, mapped_value in category_map.items():
@@ -169,51 +177,77 @@ def extract_metadata_filters(query: str):
     return gender, category, subcategory, color
 
 
-def search_fashion(query: str, alpha: float):
+def search_fashion(query: str, alpha: float, start: int = 0, end: int = 12, gender_override: str = None):
     gender, category, subcategory, color = extract_metadata_filters(query)
 
-    # Build Pinecone filter
+    # override from dropdown
+    if gender_override:
+        gender = gender_override
+
+    # --- Pinecone Filter ---
     filter = {}
+
     if gender:
         filter["gender"] = gender
+
     if category:
-        filter["articleType"] = category
+        if category in ["Footwear", "Shoes"]:
+            shoe_article_types = [
+                "Casual Shoes", "Sports Shoes", "Formal Shoes", "Training Shoes",
+                "Sneakers", "Sandals", "Slippers", "Boots", "Flip Flops"
+            ]
+            filter["articleType"] = {"$in": shoe_article_types}
+        else:
+            filter["articleType"] = category
+
     if subcategory:
         filter["subCategory"] = subcategory
+
     if color:
         filter["baseColour"] = color
 
-    print(f"🔍 Using filter: {filter}")
+    print(f"🔍 Using filter: {filter} (showing {start} to {end})")
 
-    # hybrid
     sparse = bm25.encode_queries(query)
     dense = model.encode(query).tolist()
     hdense, hsparse = hybrid_scale(dense, sparse, alpha=alpha)
 
-    # initial search
     result = index.query(
-        top_k=12,
+        top_k=end,
         vector=hdense,
         sparse_vector=hsparse,
         include_metadata=True,
         filter=filter if filter else None
     )
 
-    # fallback: if zero results with gender, relax gender
+    # fallback if no results
+    if len(result["matches"]) == 0:
+        print("⚠️ No results, retrying with alpha=0 sparse only")
+        hdense, hsparse = hybrid_scale(dense, sparse, alpha=0)
+        result = index.query(
+            top_k=end,
+            vector=hdense,
+            sparse_vector=hsparse,
+            include_metadata=True,
+            filter=filter if filter else None
+        )
+
+    # fallback if no results with gender
     if gender and len(result["matches"]) == 0:
-        print(f"⚠️ No results with gender {gender}, relaxing gender filter")
-        filter.pop("gender")
+        print(f"⚠️ No results for gender {gender}, relaxing gender filter")
+        filter.pop("gender", None)
         result = index.query(
-            top_k=12,
+            top_k=end,
             vector=hdense,
             sparse_vector=hsparse,
             include_metadata=True,
             filter=filter if filter else None
         )
 
-    # results
+    matches = result["matches"][start:end]
+
     imgs_with_captions = []
-    for r in result["matches"]:
+    for r in matches:
         idx = int(r["id"])
         img = images[idx]
         meta = r.get("metadata", {})
@@ -226,36 +260,34 @@ def search_fashion(query: str, alpha: float):
     return imgs_with_captions
 
 
-from transformers import CLIPProcessor, CLIPModel
-
-clip_model = CLIPModel.from_pretrained("openai/clip-vit-base-patch32").to(device)
-clip_processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch32")
-
 
+# this is working code block
 
 from PIL import Image, ImageOps
 import numpy as np
 
-def search_by_image(uploaded_image, alpha=0.5):
+def search_by_image(uploaded_image, alpha=0.5, start=0, end=12):
     """
-    Given a PIL image from Gradio, find visually similar products.
+    Search visually similar products with support for pagination.
     """
-    # Preprocess as CLIP expects
+    # Preprocess image for CLIP
     processed = clip_processor(images=uploaded_image, return_tensors="pt").to(device)
 
     with torch.no_grad():
         image_vec = clip_model.get_image_features(**processed)
         image_vec = image_vec.cpu().numpy().flatten().tolist()
 
-    # since your Pinecone is purely visual, we query on visual vector
+    # Query a larger top_k so you have enough to paginate
     result = index.query(
-        top_k=12,
+        top_k=end,
         vector=image_vec,
         include_metadata=True
     )
 
+    matches = result["matches"][start:end]  # slice for pagination
+
     imgs_with_captions = []
-    for r in result["matches"]:
+    for r in matches:
         idx = int(r["id"])
         img = images[idx]
         meta = r.get("metadata", {})
@@ -267,20 +299,6 @@ def search_by_image(uploaded_image, alpha=0.5):
 
     return imgs_with_captions
 
-
-
-custom_css = """
-.search-btn {
-    width: 100%;
-}
-.gr-row {
-    gap: 8px !important;   /* slightly tighter column gap */
-}
-.query-slider > div {
-    margin-bottom: 4px !important;  /* reduce space between textbox and slider */
-}
-"""
-
 # with gr.Blocks(css=custom_css) as demo:
 #     gr.Markdown("# 🛍️ Fashion Product Hybrid Search")
 
@@ -311,6 +329,18 @@ custom_css = """
 #         height="40vh"
 #     )
 
+import gradio as gr
+custom_css = """
+.search-btn {
+    width: 100%;
+}
+.gr-row {
+    gap: 8px !important;
+}
+.query-slider > div {
+    margin-bottom: 4px !important;
+}
+"""
 
 with gr.Blocks(css=custom_css) as demo:
     gr.Markdown("# 🛍️ Fashion Product Hybrid Search")
@@ -321,41 +351,73 @@ with gr.Blocks(css=custom_css) as demo:
                 label="Enter your fashion search query",
                 placeholder="Type something or leave blank to only use the image"
             )
-            alpha = gr.Slider(
-                0, 1, value=0.5,
-                label="Hybrid Weight (alpha: 0=sparse, 1=dense)"
+            alpha = gr.Slider(0, 1, value=0.5, label="Hybrid Weight (alpha: 0=sparse, 1=dense)")
+
+            gender_dropdown = gr.Dropdown(
+                ["", "Men", "Women", "Boys", "Girls", "Kids", "Unisex"],
+                label="Gender Filter (optional)"
             )
         with gr.Column(scale=1):
             image_input = gr.Image(
-                source="webcam",  # 👈 Enables webcam button
                 type="pil",
-                label="📷 Capture or Upload Image",
+                label="Upload an image (optional)",
                 height=256,
-                width=356,
-                show_label=True
+                width=356
             )
 
     search_btn = gr.Button("Search", elem_classes="search-btn")
+    gallery = gr.Gallery(label="Search Results", columns=6, height="50vh")
+    load_more_btn = gr.Button("Load More")
 
-    gallery = gr.Gallery(
-        label="Search Results",
-        columns=6,
-        height="40vh"
-    )
+    # States to track
+    search_offset = gr.State(0)
+    current_query = gr.State("")
+    current_image = gr.State(None)
+    current_gender = gr.State("")
+    shown_results = gr.State([])  # new: store the list of shown images
+
+    def unified_search(q, uploaded_image, a, offset, gender_ui):
+        start = 0
+        end = 12
 
+        gender_override = gender_ui if gender_ui else None
 
-    def unified_search(q, uploaded_image, a):
         if uploaded_image is not None:
-            return search_by_image(uploaded_image, a)
+            results = search_by_image(uploaded_image, a, start, end)
         elif q.strip() != "":
-            return search_fashion(q, a)
+            results = search_fashion(q, a, start, end, gender_override)
         else:
-            return []
+            results = []
+
+        # reset shown_results to just these first 12
+        return results, end, q, uploaded_image, gender_ui, results
 
     search_btn.click(
         unified_search,
-        inputs=[query, image_input, alpha],
-        outputs=gallery
+        inputs=[query, image_input, alpha, search_offset, gender_dropdown],
+        outputs=[gallery, search_offset, current_query, current_image, current_gender, shown_results]
+    )
+
+    def load_more_fn(a, offset, q, img, gender_ui, prev_results):
+        start = offset
+        end = offset + 12
+
+        gender_override = gender_ui if gender_ui else None
+
+        if img is not None:
+            new_results = search_by_image(img, a, start, end)
+        elif q.strip() != "":
+            new_results = search_fashion(q, a, start, end, gender_override)
+        else:
+            new_results = []
+
+        combined_results = prev_results + new_results
+        return combined_results, end, combined_results
+
+    load_more_btn.click(
+        load_more_fn,
+        inputs=[alpha, search_offset, current_query, current_image, current_gender, shown_results],
+        outputs=[gallery, search_offset, shown_results]
     )
 
     gr.Markdown("Powered by your hybrid AI search model 🚀")