Create app.py

app.py

@@ -0,0 +1,120 @@
+#Importing libraries
+import chromadb
+import time
+from transformers import CLIPModel, CLIPProcessor
+import gradio as gr
+from sklearn.metrics.pairwise import cosine_similarity
+import torch
+import numpy
+from PIL import Image
+
+#create the client
+client = chromadb.Client()
+collection = client.create_collection('image_collection')
+
+#define the model and the proccessor
+model = CLIPModel.from_pretrained('openai/clip-vit-base-patch32')
+processor = CLIPProcessor.from_pretrained  ('openai/clip-vit-base-patch32')
+
+image_paths = [
+    'img/image-01.jpg',
+    'img/image-02.jpg',
+    'img/image-03.jpeg'
+]
+
+images = [Image.open(image_path) for image_path in image_paths]
+inputs = processor(images = images, return_tensors='pt', padding=True)
+
+#Generate the embeddings
+start_time = time.time()
+with torch.no_grad():
+    image_embeddings = model.get_image_features(**inputs).numpy()
+
+image_embeddings = [embedding.tolist() for embedding in image_embeddings]
+
+end_time = time.time()
+ingestion_time = end_time - start_time
+
+collection.add(
+    embeddings = image_embeddings,
+    metadatas = [{'image':image} for image in image_paths],
+    ids = [str(i) for i in range(len(image_paths))]
+)
+
+print(f'image ingestion time {ingestion_time:.4f} seconds')
+def calculate_accuracy(image_embedding, query_embedding):
+    similarity = cosine_similarity([image_embedding], [query_embedding])[0][0]
+    return similarity
+
+def search_image(query):
+    if not query.strip():
+        return None, 'Ooopsy, you forgot to input something?, please try again ☠👻'
+
+    print(f'\nQuery: {query}')
+    start_query = time.time()
+    inputs = processor(text = query, return_tensors='pt', padding=True)
+    with torch.no_grad():
+        query_embeding = model.get_text_features(**inputs).numpy()
+
+    query_embeding = query_embeding.tolist()
+    end_query = time.time()
+    query_time = end_query - start_query
+
+    result = collection.query(query_embeddings = query_embeding, n_results=1)
+
+
+    result_image_path = result['metadatas'][0][0]['image']
+    result_image_index = int(result['ids'][0][0])
+    matched_image_embedding = image_embeddings[result_image_index]
+
+    accuracy_score = calculate_accuracy(matched_image_embedding, query_embeding[0])
+
+    result_image = Image.open(result_image_path)
+    file_name = result_image_path.split('/')[-1]
+
+    return result_image, f'Accuracy score: {accuracy_score:.4f}\nQuery Time: {query_time:.4f} seconds', file_name
+
+queries = [
+    'A group of polar bears',
+    'A famous landmark in paris',
+    'A hot pizza fresh from the oven',
+    'food',
+    'A place',
+    'A structure in Europe',
+    'Animals'
+]
+
+def populate_queries(suggested_querries):
+    return suggested_querries
+
+#defining a gradio interface
+with gr.Blocks() as gr_interface:
+    gr.Markdown('# This is an image retrieval application Made by Allan Munene🤗😁🤑')
+
+    with gr.Row():
+        with gr.Column():
+            gr.Markdown(f'***Image Ingestion Time***: {ingestion_time:.4f} seconds')
+            gr.Markdown('## Input Panel')
+            custom_query = gr.Textbox(placeholder = 'Input your query here', label='What are you looking for?')
+            with gr.Row():
+                submit_button = gr.Button('Submit Query')
+                cancel_button = gr.Button('Cancel')
+            
+            with gr.Row(elem_id='button-container'):
+                for query in queries:
+                    gr.Button(query).click(fn=lambda q=query: q, outputs=custom_query)
+
+    
+        with gr.Column():
+            gr.Markdown('Output Image')
+            Output_image = gr.Image(type='pil', label='Result_image')
+
+            accuracy = gr.Textbox(label='Performance')
+            f_name = gr.Textbox(label='File Name')
+
+        submit_button.click(fn=search_image, inputs=custom_query, outputs=[Output_image, accuracy, f_name])
+        cancel_button.click(fn=lambda: (None, ''), outputs=[Output_image, accuracy, f_name])
+        
+gr_interface.launch(share=True)
+
+