Spaces:
Sleeping
Sleeping
Update src/streamlit_app.py
Browse files- src/streamlit_app.py +58 -39
src/streamlit_app.py
CHANGED
@@ -9,7 +9,7 @@ from skimage.io import imsave
|
|
9 |
from torchvision.datasets import CIFAR10
|
10 |
import torchvision.transforms as T
|
11 |
|
12 |
-
#
|
13 |
HF_CACHE = os.path.join(tempfile.gettempdir(), "hf_cache")
|
14 |
os.makedirs(HF_CACHE, exist_ok=True)
|
15 |
os.environ["XDG_CACHE_HOME"] = HF_CACHE
|
@@ -19,12 +19,13 @@ from chromadb import PersistentClient
|
|
19 |
from chromadb.utils.data_loaders import ImageLoader
|
20 |
from chromadb.utils.embedding_functions import OpenCLIPEmbeddingFunction
|
21 |
|
22 |
-
#
|
23 |
TEMP_DIR = tempfile.gettempdir()
|
24 |
IMAGES_DIR = os.path.join(TEMP_DIR, "extracted_images")
|
25 |
DB_PATH = os.path.join(TEMP_DIR, "image_vdb")
|
26 |
os.makedirs(IMAGES_DIR, exist_ok=True)
|
27 |
|
|
|
28 |
@st.cache_resource
|
29 |
def get_chroma_collection():
|
30 |
chroma_client = PersistentClient(path=DB_PATH)
|
@@ -37,7 +38,7 @@ def get_chroma_collection():
|
|
37 |
|
38 |
image_collection = get_chroma_collection()
|
39 |
|
40 |
-
#
|
41 |
def extract_images_from_pdf(pdf_bytes):
|
42 |
pdf = fitz.open(stream=pdf_bytes, filetype="pdf")
|
43 |
saved = []
|
@@ -46,20 +47,22 @@ def extract_images_from_pdf(pdf_bytes):
|
|
46 |
base = pdf.extract_image(img[0])
|
47 |
ext = base["ext"]
|
48 |
path = os.path.join(IMAGES_DIR, f"pdf_p{i+1}_img{img[0]}.{ext}")
|
49 |
-
with open(path,"wb") as f:
|
|
|
50 |
saved.append(path)
|
51 |
return saved
|
52 |
|
|
|
53 |
def index_images(paths):
|
54 |
ids, uris = [], []
|
55 |
for path in sorted(paths):
|
56 |
-
if path.lower().endswith((".jpg",".jpeg",".png")):
|
57 |
ids.append(str(uuid.uuid4()))
|
58 |
uris.append(path)
|
59 |
if ids:
|
60 |
image_collection.add(ids=ids, uris=uris)
|
61 |
|
62 |
-
#
|
63 |
def query_similar_images(image_file, top_k=5):
|
64 |
with tempfile.NamedTemporaryFile(delete=False, suffix=".jpg") as tmp:
|
65 |
tmp.write(image_file.read())
|
@@ -68,16 +71,19 @@ def query_similar_images(image_file, top_k=5):
|
|
68 |
os.remove(tmp.name)
|
69 |
return res['uris'][0]
|
70 |
|
|
|
71 |
def search_images_by_text(text, top_k=5):
|
72 |
res = image_collection.query(query_texts=[text], n_results=top_k)
|
73 |
return res['uris'][0]
|
74 |
|
75 |
-
#
|
76 |
@st.cache_resource
|
77 |
def load_demo_cifar10(n=500):
|
78 |
dataset = CIFAR10(root=TEMP_DIR, download=True, train=True)
|
79 |
transform = T.ToPILImage()
|
80 |
saved = []
|
|
|
|
|
81 |
for i in range(min(n, len(dataset))):
|
82 |
img, label = dataset[i]
|
83 |
if not isinstance(img, Image.Image):
|
@@ -85,53 +91,66 @@ def load_demo_cifar10(n=500):
|
|
85 |
path = os.path.join(IMAGES_DIR, f"cifar10_{i}_{label}.png")
|
86 |
img.save(path)
|
87 |
saved.append(path)
|
|
|
|
|
|
|
88 |
return saved
|
89 |
|
90 |
-
#
|
91 |
-
st.title("๐ Image
|
92 |
|
93 |
-
|
|
|
94 |
|
95 |
-
if choice=="Upload PDF":
|
96 |
-
pdf = st.file_uploader("๐ค Upload PDF", type=["pdf"])
|
97 |
if pdf:
|
98 |
-
with st.spinner("Extracting..."):
|
99 |
-
imgs = extract_images_from_pdf(pdf.read())
|
100 |
-
|
|
|
101 |
st.image(imgs, width=120)
|
102 |
|
103 |
-
elif choice=="Upload Images":
|
104 |
-
imgs = st.file_uploader("๐ค Upload
|
105 |
if imgs:
|
106 |
-
|
107 |
-
|
108 |
-
|
109 |
-
|
110 |
-
|
111 |
-
|
|
|
|
|
|
|
112 |
st.image(paths, width=120)
|
113 |
|
114 |
-
elif choice=="Load CIFARโ10 Demo":
|
115 |
if st.button("๐ Load 500 CIFARโ10 Images"):
|
116 |
-
|
117 |
-
|
|
|
|
|
118 |
st.image(paths[:20], width=100)
|
119 |
|
|
|
120 |
st.divider()
|
121 |
-
st.subheader("
|
122 |
-
q = st.file_uploader("Upload a query image", type=["jpg","jpeg","png"])
|
123 |
if q:
|
124 |
-
st.image(q, caption="Query")
|
125 |
-
with st.spinner("
|
126 |
-
|
127 |
-
st.subheader("Top
|
128 |
-
for u in
|
|
|
129 |
|
130 |
st.divider()
|
131 |
-
st.subheader("๐ Text-to-Image
|
132 |
-
txt = st.text_input("
|
133 |
if txt:
|
134 |
-
with st.spinner("
|
135 |
-
|
136 |
-
st.subheader("
|
137 |
-
for u in
|
|
|
|
9 |
from torchvision.datasets import CIFAR10
|
10 |
import torchvision.transforms as T
|
11 |
|
12 |
+
# Set HuggingFace cache directory
|
13 |
HF_CACHE = os.path.join(tempfile.gettempdir(), "hf_cache")
|
14 |
os.makedirs(HF_CACHE, exist_ok=True)
|
15 |
os.environ["XDG_CACHE_HOME"] = HF_CACHE
|
|
|
19 |
from chromadb.utils.data_loaders import ImageLoader
|
20 |
from chromadb.utils.embedding_functions import OpenCLIPEmbeddingFunction
|
21 |
|
22 |
+
# Paths
|
23 |
TEMP_DIR = tempfile.gettempdir()
|
24 |
IMAGES_DIR = os.path.join(TEMP_DIR, "extracted_images")
|
25 |
DB_PATH = os.path.join(TEMP_DIR, "image_vdb")
|
26 |
os.makedirs(IMAGES_DIR, exist_ok=True)
|
27 |
|
28 |
+
# Init ChromaDB collection
|
29 |
@st.cache_resource
|
30 |
def get_chroma_collection():
|
31 |
chroma_client = PersistentClient(path=DB_PATH)
|
|
|
38 |
|
39 |
image_collection = get_chroma_collection()
|
40 |
|
41 |
+
# --- Extract images from PDF ---
|
42 |
def extract_images_from_pdf(pdf_bytes):
|
43 |
pdf = fitz.open(stream=pdf_bytes, filetype="pdf")
|
44 |
saved = []
|
|
|
47 |
base = pdf.extract_image(img[0])
|
48 |
ext = base["ext"]
|
49 |
path = os.path.join(IMAGES_DIR, f"pdf_p{i+1}_img{img[0]}.{ext}")
|
50 |
+
with open(path, "wb") as f:
|
51 |
+
f.write(base["image"])
|
52 |
saved.append(path)
|
53 |
return saved
|
54 |
|
55 |
+
# --- Index images ---
|
56 |
def index_images(paths):
|
57 |
ids, uris = [], []
|
58 |
for path in sorted(paths):
|
59 |
+
if path.lower().endswith((".jpg", ".jpeg", ".png", ".bmp", ".tiff", ".webp")):
|
60 |
ids.append(str(uuid.uuid4()))
|
61 |
uris.append(path)
|
62 |
if ids:
|
63 |
image_collection.add(ids=ids, uris=uris)
|
64 |
|
65 |
+
# --- Image-to-Image search ---
|
66 |
def query_similar_images(image_file, top_k=5):
|
67 |
with tempfile.NamedTemporaryFile(delete=False, suffix=".jpg") as tmp:
|
68 |
tmp.write(image_file.read())
|
|
|
71 |
os.remove(tmp.name)
|
72 |
return res['uris'][0]
|
73 |
|
74 |
+
# --- Text-to-Image search ---
|
75 |
def search_images_by_text(text, top_k=5):
|
76 |
res = image_collection.query(query_texts=[text], n_results=top_k)
|
77 |
return res['uris'][0]
|
78 |
|
79 |
+
# --- Load CIFAR-10 Demo Dataset (500 images) ---
|
80 |
@st.cache_resource
|
81 |
def load_demo_cifar10(n=500):
|
82 |
dataset = CIFAR10(root=TEMP_DIR, download=True, train=True)
|
83 |
transform = T.ToPILImage()
|
84 |
saved = []
|
85 |
+
|
86 |
+
progress_bar = st.progress(0)
|
87 |
for i in range(min(n, len(dataset))):
|
88 |
img, label = dataset[i]
|
89 |
if not isinstance(img, Image.Image):
|
|
|
91 |
path = os.path.join(IMAGES_DIR, f"cifar10_{i}_{label}.png")
|
92 |
img.save(path)
|
93 |
saved.append(path)
|
94 |
+
if i % 10 == 0 or i == n - 1:
|
95 |
+
progress_bar.progress((i + 1) / n)
|
96 |
+
|
97 |
return saved
|
98 |
|
99 |
+
# === UI START ===
|
100 |
+
st.title("๐ Semantic Image Search App")
|
101 |
|
102 |
+
# Step 1: Load data
|
103 |
+
choice = st.radio("๐ Select Image Source", ["Upload PDF", "Upload Images", "Load CIFARโ10 Demo"], horizontal=True)
|
104 |
|
105 |
+
if choice == "Upload PDF":
|
106 |
+
pdf = st.file_uploader("๐ค Upload PDF file", type=["pdf"])
|
107 |
if pdf:
|
108 |
+
with st.spinner("Extracting images from PDF..."):
|
109 |
+
imgs = extract_images_from_pdf(pdf.read())
|
110 |
+
index_images(imgs)
|
111 |
+
st.success(f"โ
Indexed {len(imgs)} images from PDF.")
|
112 |
st.image(imgs, width=120)
|
113 |
|
114 |
+
elif choice == "Upload Images":
|
115 |
+
imgs = st.file_uploader("๐ค Upload image files", type=["jpg", "jpeg", "png", "bmp", "tiff", "webp"], accept_multiple_files=True)
|
116 |
if imgs:
|
117 |
+
with st.spinner("Indexing uploaded images..."):
|
118 |
+
paths = []
|
119 |
+
for item in imgs:
|
120 |
+
p = os.path.join(IMAGES_DIR, item.name)
|
121 |
+
with open(p, "wb") as f:
|
122 |
+
f.write(item.read())
|
123 |
+
paths.append(p)
|
124 |
+
index_images(paths)
|
125 |
+
st.success(f"โ
{len(paths)} images indexed.")
|
126 |
st.image(paths, width=120)
|
127 |
|
128 |
+
elif choice == "Load CIFARโ10 Demo":
|
129 |
if st.button("๐ Load 500 CIFARโ10 Images"):
|
130 |
+
with st.spinner("Loading CIFARโ10 demo dataset..."):
|
131 |
+
paths = load_demo_cifar10(500)
|
132 |
+
index_images(paths)
|
133 |
+
st.success("โ
500 demo images loaded and indexed.")
|
134 |
st.image(paths[:20], width=100)
|
135 |
|
136 |
+
# Step 2: Search
|
137 |
st.divider()
|
138 |
+
st.subheader("๐ผ๏ธ Image-to-Image Search")
|
139 |
+
q = st.file_uploader("๐ท Upload a query image", type=["jpg", "jpeg", "png", "bmp", "tiff", "webp"])
|
140 |
if q:
|
141 |
+
st.image(q, caption="Query Image", width=200)
|
142 |
+
with st.spinner("Finding similar images..."):
|
143 |
+
results = query_similar_images(q, top_k=5)
|
144 |
+
st.subheader("๐ Top Matches:")
|
145 |
+
for u in results:
|
146 |
+
st.image(u, width=150)
|
147 |
|
148 |
st.divider()
|
149 |
+
st.subheader("๐ Text-to-Image Search")
|
150 |
+
txt = st.text_input("Describe what youโre looking for (e.g., 'a beach', 'a cat', 'a red truck'):")
|
151 |
if txt:
|
152 |
+
with st.spinner("Finding images by semantic similarity..."):
|
153 |
+
results = search_images_by_text(txt, top_k=5)
|
154 |
+
st.subheader("๐ Semantic Matches:")
|
155 |
+
for u in results:
|
156 |
+
st.image(u, width=150)
|