import streamlit as st import open_clip import torch import requests from PIL import Image from io import BytesIO import time import numpy as np from ultralytics import YOLO import chromadb from transformers import pipeline from sklearn.metrics.pairwise import cosine_similarity # Load segmentation model segmenter = pipeline(model="mattmdjaga/segformer_b2_clothes") # Load CLIP model and tokenizer @st.cache_resource def load_clip_model(): model, preprocess_train, preprocess_val = open_clip.create_model_and_transforms('hf-hub:Marqo/marqo-fashionSigLIP') tokenizer = open_clip.get_tokenizer('hf-hub:Marqo/marqo-fashionSigLIP') device = torch.device("cuda" if torch.cuda.is_available() else "cpu") model.to(device) return model, preprocess_val, tokenizer, device clip_model, preprocess_val, tokenizer, device = load_clip_model() # Load YOLOv8 model @st.cache_resource def load_yolo_model(): return YOLO("./best.pt") yolo_model = load_yolo_model() # Load chromaDB client = chromadb.PersistentClient(path="./clothesDB_Test") #collection = client.get_collection(name="clothes_items_ver3") collection = client.get_collection(name="category_upper") # Helper functions def load_image_from_url(url, max_retries=3): for attempt in range(max_retries): try: response = requests.get(url, timeout=10) response.raise_for_status() img = Image.open(BytesIO(response.content)).convert('RGB') return img except (requests.RequestException, Image.UnidentifiedImageError) as e: if attempt < max_retries - 1: time.sleep(1) else: return None def get_image_embedding(image): image_tensor = preprocess_val(image).unsqueeze(0).to(device) with torch.no_grad(): image_features = clip_model.encode_image(image_tensor) image_features /= image_features.norm(dim=-1, keepdim=True) return image_features.cpu().numpy() def segment_clothing(img, clothes=["Hat", "Upper-clothes", "Skirt", "Pants", "Dress", "Belt", "Left-shoe", "Right-shoe", "Scarf"]): # Segment image segments = segmenter(img) # Create list of masks mask_list = [] detected_categories = [] for s in segments: if s['label'] in clothes: mask_list.append(s['mask']) detected_categories.append(s['label']) # Store detected categories # Paste all masks on top of each other final_mask = np.zeros_like(np.array(img)[:, :, 0]) # Initialize mask for mask in mask_list: current_mask = np.array(mask) final_mask = np.maximum(final_mask, current_mask) # Use maximum to combine masks # Convert final mask from np array to PIL image final_mask = Image.fromarray(final_mask.astype(np.uint8) * 255) # Convert to binary mask # Apply mask to original image img_with_alpha = img.convert("RGBA") # Ensure the image has an alpha channel img_with_alpha.putalpha(final_mask) return img_with_alpha.convert("RGB"), final_mask, detected_categories # Return detected categories #def find_similar_images(query_embedding, collection, top_k=5): # all_embeddings = collection.get(include=['embeddings'])['embeddings'] # database_embeddings = np.array(all_embeddings) # similarities = np.dot(database_embeddings, query_embedding.T).squeeze() # top_indices = np.argsort(similarities)[::-1][:top_k] # # all_data = collection.get(include=['metadatas'])['metadatas'] # top_metadatas = [all_data[idx] for idx in top_indices] # results = [] # for idx, metadata in enumerate(top_metadatas): # results.append({ # 'info': metadata, # 'similarity': similarities[top_indices[idx]] # }) # return results def find_similar_images(query_embedding, collection, top_k=5): # 모든 임베딩을 가져옴 all_embeddings = collection.get(include=['embeddings'])['embeddings'] database_embeddings = np.array(all_embeddings) # 유사도 계산 similarities = cosine_similarity(database_embeddings, query_embedding.reshape(1, -1)).squeeze() top_indices = np.argsort(similarities)[::-1][:top_k] # 메타데이터 가져옴 all_data = collection.get(include=['metadatas'])['metadatas'] top_metadatas = [all_data[idx] for idx in top_indices] results = [] for idx, metadata in enumerate(top_metadatas): # 이미지 URLs 필드가 쉼표로 구분된 문자열로 저장된 경우, 이를 리스트로 변환 image_urls = metadata['image_url'].split(',') # 첫 번째 이미지를 대표 이미지로 사용 representative_image_url = image_urls[0] if image_urls else None results.append({ 'info': metadata, 'similarity': similarities[top_indices[idx]], 'image_url': representative_image_url # 첫 번째 이미지 URL 사용 }) return results # 세션 상태 초기화 if 'step' not in st.session_state: st.session_state.step = 'input' if 'query_image_url' not in st.session_state: st.session_state.query_image_url = '' if 'detections' not in st.session_state: st.session_state.detections = [] if 'segmented_image' not in st.session_state: # Add segmented_image to session state st.session_state.segmented_image = None if 'selected_category' not in st.session_state: st.session_state.selected_category = None # Streamlit app st.title("Advanced Fashion Search App") # 단계별 처리 if st.session_state.step == 'input': st.session_state.query_image_url = st.text_input("Enter image URL:", st.session_state.query_image_url) if st.button("Detect Clothing"): if st.session_state.query_image_url: query_image = load_image_from_url(st.session_state.query_image_url) if query_image is not None: st.session_state.query_image = query_image # Perform segmentation segmented_image, final_mask, detected_categories = segment_clothing(query_image) st.session_state.segmented_image = segmented_image # Store segmented image in session state st.session_state.detections = detected_categories # Store detected categories st.image(segmented_image, caption="Segmented Image", use_column_width=True) if st.session_state.detections: st.session_state.step = 'select_category' else: st.warning("No clothing items detected in the image.") else: st.error("Failed to load the image. Please try another URL.") else: st.warning("Please enter an image URL.") elif st.session_state.step == 'select_category': st.image(st.session_state.segmented_image, caption="Segmented Image with Detected Categories", use_column_width=True) st.subheader("Detected Clothing Categories:") if st.session_state.detections: selected_category = st.selectbox("Select a category to search:", st.session_state.detections) if st.button("Search Similar Items"): st.session_state.selected_category = selected_category st.session_state.step = 'show_results' else: st.warning("No categories detected.") elif st.session_state.step == 'show_results': original_image = st.session_state.query_image.convert("RGB") # Convert to RGB before displaying st.image(original_image, caption="Original Image", use_column_width=True) # Get the embedding of the segmented image query_embedding = get_image_embedding(st.session_state.segmented_image) # Use the segmented image from session state similar_images = find_similar_images(query_embedding, collection) st.subheader("Similar Items:") for img in similar_images: col1, col2 = st.columns(2) with col1: st.image(img['image_url'], use_column_width=True) with col2: st.write(f"Name: {img['info']['name']}") st.write(f"Brand: {img['info']['brand']}") category = img['info'].get('category') if category: st.write(f"Category: {category}") st.write(f"Price: {img['info']['price']}") st.write(f"Discount: {img['info']['discount']}%") st.write(f"Similarity: {img['similarity']:.2f}") if st.button("Start New Search"): st.session_state.step = 'input' st.session_state.query_image_url = '' st.session_state.detections = [] st.session_state.segmented_image = None # Reset segmented_image