Spaces:
Build error
Build error
File size: 3,107 Bytes
17c4cbe ed481cf 17c4cbe ed481cf 17c4cbe 459cd3a ed481cf 44485b6 17c4cbe 459cd3a 17c4cbe 459cd3a 17c4cbe cad54b2 d121302 44485b6 |
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 |
from cProfile import label
from turtle import title
import numpy as np
import gradio as gr
import pickle
from skimage import io
from scipy.spatial import distance
# all the images name in a list
images = [line.strip() for line in open("holidays_images.dat","r")]
# all the query image names in a list
query_images = []
for line in open("holidays_images.dat","r"):
imname=line.strip()
imno=int(imname[:-len(".jpg")])
if imno%100==0:
query_images.append(imname)
with open('saved_cnn.pkl', 'rb') as f:
cnn_embeddings = pickle.load(f)
with open('saved_bovw.pkl', 'rb') as f:
bovw_embeddings = pickle.load(f)
with open('saved_naive.pkl', 'rb') as f:
naive_embeddings = pickle.load(f)
def similarity_all(query_image_name, embeddings, metric):
querry_embedding = embeddings[query_image_name]
scores = {image_name : metric(querry_embedding, embeddings[image_name]) for image_name in images}
return scores
def euclidean_similarity_score(query_embedding, target_embedding):
return np.linalg.norm(query_embedding-target_embedding)
def cosine_similarity_score(query_embedding, target_embedding):
return distance.cosine(np.reshape(query_embedding, -1), np.reshape(target_embedding, -1))
def retrieve(query_image_name, embeddings_type, metric_type):
if embeddings_type == 'MobileNetV2' :
embeddings = cnn_embeddings
elif embeddings_type == 'BoVW' :
embeddings = bovw_embeddings
else :
embeddings = naive_embeddings
if metric_type == 'Euclidean' :
metric = euclidean_similarity_score
else :
metric = cosine_similarity_score
scores = similarity_all(query_image_name, embeddings, metric)
top = sorted(scores, key=scores.get)[:11]
return io.imread('smallholidays/'+top[0]), [io.imread('smallholidays/'+img) for img in top[1:]]
input_button = gr.inputs.Dropdown(query_images, label='Choice of the query image')
embeddings_selection = gr.inputs.Radio(['MobileNetV2', 'BoVW', 'Baseline'], label='Embeddings')
metric_selection = gr.inputs.Radio(['Euclidean', 'Cosine'], label='Similarity Metric')
retrieved_images = gr.outputs.Carousel(["image"], label='Retrieved images')
description = "This is a demo of the content-based image retrieval system developed as part of the IR course project, 2022. The indexed dataset is [INRIA Holidays](https://lear.inrialpes.fr/~jegou/data.php). \n\nSeveral image embeddings can be used :\n \n-**MobileNetV2** : feature extraction is performed using a MobileNet architecture trained on ImageNet.\n\n-**BoVW (Bag of Visual Words)** : embedding is the BoVW histogram using color histogram as a descriptor.\n\n-**Baseline** : basic descriptor that uses pixel values of the downsized images."
iface = gr.Interface(fn=retrieve,
inputs=[input_button, embeddings_selection, metric_selection],
outputs=[gr.outputs.Image(label='Query image'), retrieved_images],
title='Image Retrieval on INRIA Holidays',
description=description)
iface.launch() |