First Commit
Browse files
app.py
ADDED
@@ -0,0 +1,139 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import requests
|
3 |
+
from tqdm import tqdm
|
4 |
+
from datasets import load_dataset
|
5 |
+
import numpy as np
|
6 |
+
import tensorflow as tf
|
7 |
+
from tensorflow.keras.applications.resnet50 import ResNet50, preprocess_input
|
8 |
+
from tensorflow.keras.preprocessing import image
|
9 |
+
from sklearn.neighbors import NearestNeighbors
|
10 |
+
import joblib
|
11 |
+
from PIL import UnidentifiedImageError, Image
|
12 |
+
import gradio as gr
|
13 |
+
import matplotlib.pyplot as plt
|
14 |
+
|
15 |
+
# Ensure TensorFlow uses GPU
|
16 |
+
print("Num GPUs Available: ", len(tf.config.list_physical_devices('GPU')))
|
17 |
+
assert len(tf.config.list_physical_devices('GPU')) > 0, "No GPU available!"
|
18 |
+
|
19 |
+
# Load the dataset
|
20 |
+
dataset = load_dataset("thefcraft/civitai-stable-diffusion-337k")
|
21 |
+
|
22 |
+
# Take a subset of the dataset
|
23 |
+
subset_size = 10000
|
24 |
+
dataset_subset = dataset['train'].shuffle(seed=42).select(range(subset_size))
|
25 |
+
|
26 |
+
# Directory to save images
|
27 |
+
image_dir = 'civitai_images'
|
28 |
+
os.makedirs(image_dir, exist_ok=True)
|
29 |
+
|
30 |
+
# Load the ResNet50 model pretrained on ImageNet
|
31 |
+
with tf.device('/GPU:0'):
|
32 |
+
model = ResNet50(weights='imagenet', include_top=False, pooling='avg')
|
33 |
+
|
34 |
+
# Function to extract features
|
35 |
+
def extract_features(img_path, model):
|
36 |
+
img = image.load_img(img_path, target_size=(224, 224))
|
37 |
+
img_array = image.img_to_array(img)
|
38 |
+
img_array = np.expand_dims(img_array, axis=0)
|
39 |
+
img_array = preprocess_input(img_array)
|
40 |
+
with tf.device('/GPU:0'):
|
41 |
+
features = model.predict(img_array)
|
42 |
+
return features.flatten()
|
43 |
+
|
44 |
+
# Extract features for a sample of images
|
45 |
+
features = []
|
46 |
+
image_paths = []
|
47 |
+
model_names = []
|
48 |
+
|
49 |
+
for sample in tqdm(dataset_subset):
|
50 |
+
img_url = sample['url'] # Adjust based on the correct column name
|
51 |
+
model_name = sample['Model'] # Adjust based on the correct column name
|
52 |
+
img_path = os.path.join(image_dir, os.path.basename(img_url))
|
53 |
+
|
54 |
+
# Download the image
|
55 |
+
try:
|
56 |
+
response = requests.get(img_url)
|
57 |
+
response.raise_for_status() # Check if the download was successful
|
58 |
+
|
59 |
+
if 'image' not in response.headers['Content-Type']:
|
60 |
+
raise ValueError("URL does not contain an image")
|
61 |
+
|
62 |
+
with open(img_path, 'wb') as f:
|
63 |
+
f.write(response.content)
|
64 |
+
|
65 |
+
# Extract features
|
66 |
+
try:
|
67 |
+
img_features = extract_features(img_path, model)
|
68 |
+
features.append(img_features)
|
69 |
+
image_paths.append(img_path)
|
70 |
+
model_names.append(model_name)
|
71 |
+
except UnidentifiedImageError:
|
72 |
+
print(f"UnidentifiedImageError: Skipping file {img_path}")
|
73 |
+
os.remove(img_path)
|
74 |
+
|
75 |
+
except requests.exceptions.RequestException as e:
|
76 |
+
print(f"RequestException: Failed to download {img_url} - {e}")
|
77 |
+
|
78 |
+
# Convert features to numpy array
|
79 |
+
features = np.array(features)
|
80 |
+
|
81 |
+
# Build the NearestNeighbors model
|
82 |
+
nbrs = NearestNeighbors(n_neighbors=5, algorithm='ball_tree').fit(features)
|
83 |
+
|
84 |
+
# Save the model and features
|
85 |
+
joblib.dump(nbrs, 'nearest_neighbors_model.pkl')
|
86 |
+
np.save('image_features.npy', features)
|
87 |
+
np.save('image_paths.npy', image_paths)
|
88 |
+
np.save('model_names.npy', model_names)
|
89 |
+
|
90 |
+
# Load the NearestNeighbors model and features
|
91 |
+
nbrs = joblib.load('nearest_neighbors_model.pkl')
|
92 |
+
features = np.load('image_features.npy')
|
93 |
+
image_paths = np.load('image_paths.npy', allow_pickle=True)
|
94 |
+
model_names = np.load('model_names.npy', allow_pickle=True)
|
95 |
+
|
96 |
+
# Function to get recommendations
|
97 |
+
def get_recommendations(img_path, model, nbrs, image_paths, model_names, n_neighbors=5):
|
98 |
+
img_features = extract_features(img_path, model)
|
99 |
+
distances, indices = nbrs.kneighbors([img_features])
|
100 |
+
|
101 |
+
recommended_images = [image_paths[idx] for idx in indices.flatten()]
|
102 |
+
recommended_model_names = [model_names[idx] for idx in indices.flatten()]
|
103 |
+
recommended_distances = distances.flatten()
|
104 |
+
|
105 |
+
return recommended_images, recommended_model_names, recommended_distances
|
106 |
+
|
107 |
+
def recommend(image):
|
108 |
+
# Save uploaded image to a path
|
109 |
+
image_path = "uploaded_image.jpg"
|
110 |
+
image.save(image_path)
|
111 |
+
|
112 |
+
recommended_images, recommended_model_names, recommended_distances = get_recommendations(image_path, model, nbrs, image_paths, model_names)
|
113 |
+
result = list(zip(recommended_images, recommended_model_names, recommended_distances))
|
114 |
+
|
115 |
+
# Display images with matplotlib
|
116 |
+
display_images(recommended_images, recommended_model_names, recommended_distances)
|
117 |
+
|
118 |
+
return result
|
119 |
+
|
120 |
+
def display_images(image_paths, model_names, distances):
|
121 |
+
plt.figure(figsize=(20, 10))
|
122 |
+
for i, (img_path, model_name, distance) in enumerate(zip(image_paths, model_names, distances)):
|
123 |
+
img = Image.open(img_path)
|
124 |
+
plt.subplot(1, len(image_paths), i+1)
|
125 |
+
plt.imshow(img)
|
126 |
+
plt.title(f'{model_name}\nDistance: {distance:.2f}', fontsize=12)
|
127 |
+
plt.axis('off')
|
128 |
+
plt.show()
|
129 |
+
|
130 |
+
# Gradio interface
|
131 |
+
interface = gr.Interface(
|
132 |
+
fn=recommend,
|
133 |
+
inputs=gr.inputs.Image(type="pil"),
|
134 |
+
outputs="text", # Outputs the list of recommended images, models, and distances
|
135 |
+
title="Image Recommendation System",
|
136 |
+
description="Upload an image and get 5 recommended similar images with model names and distances."
|
137 |
+
)
|
138 |
+
|
139 |
+
interface.launch()
|