Spaces:
Sleeping
Sleeping
File size: 10,510 Bytes
520da1b | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 | import os
import sys
import json
import numpy as np
import faiss
from tqdm import tqdm
from utils.model_loader import ModelLoader
# Add the project root directory to Python path
project_root = os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
sys.path.append(project_root)
from typing import Optional, List, Dict, Union
class ImageRecommender:
def __init__(self, embed_model_path: str):
self.model = ModelLoader.get_model(embed_model_path)
self.index = None
self.image_paths = []
self.image_data = []
self.icon_indices = [] # Store indices of icon images
self.clipart_indices = [] # Store indices of clipart images
def create_index(self, image_list_path: str, image_resource_path: str):
"""Create FAISS index from image embeddings"""
# Read image paths
with open(image_list_path, 'r') as f:
self.image_paths = [line.strip().split(',') for line in f.readlines()]
# Load and process each image's data
embeddings = []
for idx, (image_path, json_path) in enumerate(tqdm(self.image_paths)):
# Construct full path to the image's JSON data
#json_path = os.path.join(image_resource_path, f'results/{idx}.json')
if not os.path.exists(json_path):
continue
try:
with open(json_path, 'r') as f:
try:
image_info = json.load(f)
except json.JSONDecodeError as e:
print(f"Error decoding JSON at index {idx}: {str(e)}")
continue
# Combine multiple fields for better semantic representation
semantic_text = ""
try:
image_content = image_info.get('image_content', '')
topic = image_info.get('topic', '')
explanation = image_info.get('explanation', '')
if image_content:
semantic_text += f"{image_content}"
if topic:
semantic_text += f". {topic}"
if explanation:
semantic_text += f". {explanation}"
if not semantic_text:
semantic_text = "Image without description"
print(f"Warning: Missing semantic information for image at index {idx}")
except Exception as e:
semantic_text = "Image without description"
print(f"Error processing semantic text at index {idx}: {str(e)}")
# Generate embedding
try:
embedding = self.model.encode(semantic_text)
embeddings.append(embedding)
self.image_data.append(image_info)
# Store indices based on image type
if image_info.get('icon_or_clipart') == 'icon':
self.icon_indices.append(len(embeddings) - 1)
elif image_info.get('icon_or_clipart') == 'clipart':
self.clipart_indices.append(len(embeddings) - 1)
except Exception as e:
print(f"Error generating embedding at index {idx}: {str(e)}")
continue
except Exception as e:
print(f"Error processing image at index {idx}: {str(e)}")
continue
self.image_paths = [image_path for image_path, _ in self.image_paths]
embeddings = np.array(embeddings).astype('float32')
# Create and train FAISS index
dimension = embeddings.shape[1]
self.index = faiss.IndexFlatL2(dimension)
self.index.add(embeddings)
def save_index(self, index_path, data_path):
"""Save the FAISS index and associated data"""
faiss.write_index(self.index, index_path)
# Save image data and indices
with open(data_path, 'w') as f:
json.dump({
'paths': self.image_paths,
'data': self.image_data,
'icon_indices': self.icon_indices,
'clipart_indices': self.clipart_indices
}, f)
def load_index(self, index_path, data_path):
"""Load the FAISS index and associated data"""
self.index = faiss.read_index(index_path)
with open(data_path, 'r') as f:
data = json.load(f)
self.image_paths = data['paths']
self.image_data = data['data']
self.icon_indices = data['icon_indices']
self.clipart_indices = data['clipart_indices']
def search(self, query_text: str, new_index = None, new_data = None, top_k: int = 5, image_type: Optional[str] = None) -> List[Dict]:
"""
Search for similar images based on query text
Args:
query_text: The text query to search for
new_index: Optional new FAISS index to search in addition
new_data: Optional new data associated with new_index
top_k: Number of results to return
image_type: Optional filter for image type ('icon' or 'clipart')
Returns:
List of dictionaries containing image information and similarity scores
"""
if self.index is None:
raise ValueError("Index not loaded. Please load the index first.")
# Generate query embedding
query_embedding = self.model.encode(query_text)
query_embedding = np.array([query_embedding]).astype('float32')
results = []
# 搜索旧索引
# Determine which indices to search in
if image_type == 'icon':
search_indices = self.icon_indices
elif image_type == 'clipart':
search_indices = self.clipart_indices
else:
search_indices = None
if search_indices:
# Create a subset index for the specific image type
subset_index = faiss.IndexFlatL2(self.index.d)
subset_index.add(self.index.reconstruct_n(0, self.index.ntotal)[search_indices])
# Search in the subset index
distances, indices = subset_index.search(query_embedding, top_k)
# Map back to original indices
indices = [search_indices[i] for i in indices[0]]
distances = distances[0]
else:
# Search in the full index
distances, indices = self.index.search(query_embedding, top_k)
indices = indices[0]
distances = distances[0]
# 添加旧索引结果
for idx, distance in zip(indices, distances):
if idx < len(self.image_paths): # Ensure index is valid
results.append({
'image_path': self.image_paths[idx],
'image_data': self.image_data[idx],
'distance': float(distance)
})
# 如果是icon类型且有新索引,搜索新索引
if image_type == 'icon' and new_index is not None and new_data is not None:
new_distances, new_indices = new_index.search(query_embedding, top_k)
new_indices = new_indices[0]
new_distances = new_distances[0]
# 添加新索引结果
for idx, distance in zip(new_indices, new_distances):
if idx < len(new_data['index']):
data = new_data['index'][str(idx)]
# print("data: ", data)
results.append({
'image_path': data["path"],
'image_data': data["data"],
'distance': float(distance) + 0.1
})
# 按距离排序并返回前top_k个结果
results.sort(key=lambda x: x['distance'])
return results[:top_k]
def main(image_list_path: str = None,
image_resource_path: str = None,
index_path: str = None,
data_path: str = None,
embed_model_path: str = None,
force: bool = False):
"""
Create and save the image index
Args:
image_list_path: Path to the file containing list of image paths
image_resource_path: Path to the directory containing image resources
index_path: Path to save the FAISS index
data_path: Path to save the image data
embed_model_path: Path to the sentence embedding model
force: Whether to force rebuild the index if it exists
"""
# Check if index exists and handle force flag
if os.path.exists(index_path) and not force:
print(f"Index file {index_path} already exists. Use --force to rebuild.")
return 0
recommender = ImageRecommender(embed_model_path)
recommender.create_index(image_list_path, image_resource_path)
recommender.save_index(index_path, data_path)
print("Index built successfully!")
return 0
if __name__ == '__main__':
import argparse
parser = argparse.ArgumentParser(description='Build image index using FAISS and SentenceTransformer')
parser.add_argument('--image_list_path', type=str, required=True, help='Path to the file containing list of image paths')
parser.add_argument('--image_resource_path', type=str, required=True, help='Path to the directory containing image resources')
parser.add_argument('--index_path', type=str, required=True, help='Path to save the FAISS index')
parser.add_argument('--data_path', type=str, required=True, help='Path to save the image data')
parser.add_argument('--embed_model_path', type=str, required=True, help='Path to the sentence embedding model')
parser.add_argument('--force', action='store_true', help='Force rebuild even if index exists')
args = parser.parse_args()
main(
image_list_path=args.image_list_path,
image_resource_path=args.image_resource_path,
index_path=args.index_path,
data_path=args.data_path,
embed_model_path=args.embed_model_path,
force=args.force
)
|