Upload app.py
Browse files
app.py
CHANGED
@@ -12,7 +12,6 @@ from transformers import (
|
|
12 |
)
|
13 |
from datasets import load_dataset
|
14 |
|
15 |
-
# Загрузка датасета и моделей
|
16 |
wikiart_dataset = load_dataset("huggan/wikiart", split="train")
|
17 |
device = torch.device("cuda" if torch.cuda.is_available() else "mps" if torch.backends.mps.is_available() else "cpu")
|
18 |
|
@@ -22,11 +21,9 @@ blip_model = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image
|
|
22 |
clip_model = CLIPModel.from_pretrained("openai/clip-vit-base-patch32").to(device).eval()
|
23 |
clip_processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch32")
|
24 |
|
25 |
-
# Загрузка FAISS индексов
|
26 |
image_index = faiss.read_index("image_index.faiss")
|
27 |
text_index = faiss.read_index("text_index.faiss")
|
28 |
|
29 |
-
# Генерация описания через BLIP
|
30 |
def generate_caption(image: Image.Image):
|
31 |
inputs = blip_processor(image, return_tensors="pt").to(device)
|
32 |
with torch.no_grad():
|
@@ -34,7 +31,6 @@ def generate_caption(image: Image.Image):
|
|
34 |
caption = blip_processor.decode(caption_ids[0], skip_special_tokens=True)
|
35 |
return caption
|
36 |
|
37 |
-
# Получение CLIP эмбеддинга по тексту
|
38 |
def get_clip_text_embedding(text):
|
39 |
inputs = clip_processor(text=[text], return_tensors="pt", padding=True).to(device)
|
40 |
with torch.no_grad():
|
@@ -43,7 +39,6 @@ def get_clip_text_embedding(text):
|
|
43 |
faiss.normalize_L2(features)
|
44 |
return features
|
45 |
|
46 |
-
# Получение CLIP эмбеддинга по изображению
|
47 |
def get_clip_image_embedding(image):
|
48 |
inputs = clip_processor(images=image, return_tensors="pt").to(device)
|
49 |
with torch.no_grad():
|
@@ -52,13 +47,12 @@ def get_clip_image_embedding(image):
|
|
52 |
faiss.normalize_L2(features)
|
53 |
return features
|
54 |
|
55 |
-
# Получение похожих изображений по эмбеддингу
|
56 |
def get_results_with_images(embedding, index, top_k=2):
|
57 |
D, I = index.search(embedding, top_k)
|
58 |
results = []
|
59 |
for idx in I[0]:
|
60 |
try:
|
61 |
-
item = wikiart_dataset[idx]
|
62 |
img = item["image"]
|
63 |
title = item.get("title", "Untitled")
|
64 |
artist = item.get("artist", "Unknown")
|
@@ -79,7 +73,6 @@ def search_similar_images(image: Image.Image):
|
|
79 |
|
80 |
return caption, text_results, image_results
|
81 |
|
82 |
-
# Интерфейс Gradio
|
83 |
demo = gr.Interface(
|
84 |
fn=search_similar_images,
|
85 |
inputs=gr.Image(label="Загрузите изображение", type="pil"),
|
|
|
12 |
)
|
13 |
from datasets import load_dataset
|
14 |
|
|
|
15 |
wikiart_dataset = load_dataset("huggan/wikiart", split="train")
|
16 |
device = torch.device("cuda" if torch.cuda.is_available() else "mps" if torch.backends.mps.is_available() else "cpu")
|
17 |
|
|
|
21 |
clip_model = CLIPModel.from_pretrained("openai/clip-vit-base-patch32").to(device).eval()
|
22 |
clip_processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch32")
|
23 |
|
|
|
24 |
image_index = faiss.read_index("image_index.faiss")
|
25 |
text_index = faiss.read_index("text_index.faiss")
|
26 |
|
|
|
27 |
def generate_caption(image: Image.Image):
|
28 |
inputs = blip_processor(image, return_tensors="pt").to(device)
|
29 |
with torch.no_grad():
|
|
|
31 |
caption = blip_processor.decode(caption_ids[0], skip_special_tokens=True)
|
32 |
return caption
|
33 |
|
|
|
34 |
def get_clip_text_embedding(text):
|
35 |
inputs = clip_processor(text=[text], return_tensors="pt", padding=True).to(device)
|
36 |
with torch.no_grad():
|
|
|
39 |
faiss.normalize_L2(features)
|
40 |
return features
|
41 |
|
|
|
42 |
def get_clip_image_embedding(image):
|
43 |
inputs = clip_processor(images=image, return_tensors="pt").to(device)
|
44 |
with torch.no_grad():
|
|
|
47 |
faiss.normalize_L2(features)
|
48 |
return features
|
49 |
|
|
|
50 |
def get_results_with_images(embedding, index, top_k=2):
|
51 |
D, I = index.search(embedding, top_k)
|
52 |
results = []
|
53 |
for idx in I[0]:
|
54 |
try:
|
55 |
+
item = wikiart_dataset[int(idx)]
|
56 |
img = item["image"]
|
57 |
title = item.get("title", "Untitled")
|
58 |
artist = item.get("artist", "Unknown")
|
|
|
73 |
|
74 |
return caption, text_results, image_results
|
75 |
|
|
|
76 |
demo = gr.Interface(
|
77 |
fn=search_similar_images,
|
78 |
inputs=gr.Image(label="Загрузите изображение", type="pil"),
|