Update app.py
Browse files
app.py
CHANGED
@@ -12,10 +12,6 @@ from inference_sdk import InferenceHTTPClient
|
|
12 |
import matplotlib.pyplot as plt
|
13 |
import base64
|
14 |
|
15 |
-
# ์ค๋ฅ ์ฒ๋ฆฌ๋ฅผ ์ํ ์์ธ ํด๋์ค ์ ์
|
16 |
-
class APIError(Exception):
|
17 |
-
pass
|
18 |
-
|
19 |
# Load model and tokenizer
|
20 |
@st.cache_resource
|
21 |
def load_model():
|
@@ -26,12 +22,62 @@ def load_model():
|
|
26 |
|
27 |
model, preprocess_val, tokenizer, device = load_model()
|
28 |
|
29 |
-
#
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
35 |
|
36 |
# Streamlit app
|
37 |
st.title("Fashion Search App with Segmentation")
|
@@ -42,127 +88,6 @@ api_key = st.text_input("Enter your Roboflow API Key", type="password")
|
|
42 |
if api_key:
|
43 |
CLIENT = setup_roboflow_client(api_key)
|
44 |
|
45 |
-
def segment_image(image_path):
|
46 |
-
try:
|
47 |
-
# ์ด๋ฏธ์ง ํ์ผ ์ฝ๊ธฐ
|
48 |
-
with open(image_path, "rb") as image_file:
|
49 |
-
image_data = image_file.read()
|
50 |
-
|
51 |
-
# ์ด๋ฏธ์ง๋ฅผ base64๋ก ์ธ์ฝ๋ฉ
|
52 |
-
encoded_image = base64.b64encode(image_data).decode('utf-8')
|
53 |
-
|
54 |
-
# ์๋ณธ ์ด๋ฏธ์ง ๋ก๋
|
55 |
-
image = cv2.imread(image_path)
|
56 |
-
image = cv2.resize(image, (800, 600))
|
57 |
-
mask = np.zeros(image.shape, dtype=np.uint8)
|
58 |
-
|
59 |
-
try:
|
60 |
-
# Roboflow API ํธ์ถ
|
61 |
-
results = CLIENT.infer(encoded_image, model_id="closet/1")
|
62 |
-
except Exception as api_error:
|
63 |
-
st.error(f"API Error: {str(api_error)}")
|
64 |
-
return Image.fromarray(cv2.cvtColor(image, cv2.COLOR_BGR2RGB))
|
65 |
-
|
66 |
-
if 'predictions' in results:
|
67 |
-
for prediction in results['predictions']:
|
68 |
-
points = prediction['points']
|
69 |
-
pts = np.array([[p['x'], p['y']] for p in points], np.int32)
|
70 |
-
scale_x = image.shape[1] / results['image']['width']
|
71 |
-
scale_y = image.shape[0] / results['image']['height']
|
72 |
-
pts = pts * [scale_x, scale_y]
|
73 |
-
pts = pts.astype(np.int32)
|
74 |
-
pts = pts.reshape((-1, 1, 2))
|
75 |
-
cv2.fillPoly(mask, [pts], color=(255, 255, 255)) # White mask
|
76 |
-
|
77 |
-
segmented_image = cv2.bitwise_and(image, mask)
|
78 |
-
else:
|
79 |
-
st.warning("No predictions found in the image. Returning original image.")
|
80 |
-
segmented_image = image
|
81 |
-
|
82 |
-
return Image.fromarray(cv2.cvtColor(segmented_image, cv2.COLOR_BGR2RGB))
|
83 |
-
except Exception as e:
|
84 |
-
st.error(f"Error in segmentation: {str(e)}")
|
85 |
-
# ์๋ณธ ์ด๋ฏธ์ง๋ฅผ ๋ค์ ์ฝ์ด ๋ฐํ
|
86 |
-
return Image.open(image_path)
|
87 |
-
def get_image_embedding(image):
|
88 |
-
image_tensor = preprocess_val(image).unsqueeze(0).to(device)
|
89 |
-
with torch.no_grad():
|
90 |
-
image_features = model.encode_image(image_tensor)
|
91 |
-
image_features /= image_features.norm(dim=-1, keepdim=True)
|
92 |
-
return image_features.cpu().numpy()
|
93 |
-
|
94 |
-
# Load and process data
|
95 |
-
@st.cache_data
|
96 |
-
def load_data():
|
97 |
-
with open('musinsa-final.json', 'r', encoding='utf-8') as f:
|
98 |
-
return json.load(f)
|
99 |
-
|
100 |
-
data = load_data()
|
101 |
-
|
102 |
-
# Process database with segmentation
|
103 |
-
@st.cache_data
|
104 |
-
def download_and_process_image(image_url):
|
105 |
-
try:
|
106 |
-
response = requests.get(image_url)
|
107 |
-
response.raise_for_status() # Raises an HTTPError for bad responses
|
108 |
-
image = Image.open(BytesIO(response.content))
|
109 |
-
|
110 |
-
# Convert image to RGB mode if it's in RGBA mode
|
111 |
-
if image.mode == 'RGBA':
|
112 |
-
image = image.convert('RGB')
|
113 |
-
|
114 |
-
return image
|
115 |
-
except requests.RequestException as e:
|
116 |
-
st.error(f"Error downloading image: {e}")
|
117 |
-
return None
|
118 |
-
except Exception as e:
|
119 |
-
st.error(f"Error processing image: {e}")
|
120 |
-
return None
|
121 |
-
|
122 |
-
def process_database():
|
123 |
-
database_embeddings = []
|
124 |
-
database_info = []
|
125 |
-
for item in data:
|
126 |
-
image_url = item['์ด๋ฏธ์ง ๋งํฌ'][0]
|
127 |
-
product_id = item.get('\ufeff์ํ ID') or item.get('์ํ ID')
|
128 |
-
|
129 |
-
image = download_and_process_image(image_url)
|
130 |
-
if image is None:
|
131 |
-
continue
|
132 |
-
|
133 |
-
# Save the image temporarily
|
134 |
-
temp_path = f"temp_{product_id}.jpg"
|
135 |
-
image.save(temp_path, 'JPEG')
|
136 |
-
|
137 |
-
segmented_image = segment_image(temp_path)
|
138 |
-
embedding = get_image_embedding(segmented_image)
|
139 |
-
|
140 |
-
database_embeddings.append(embedding)
|
141 |
-
database_info.append({
|
142 |
-
'id': product_id,
|
143 |
-
'category': item['์นดํ
๊ณ ๋ฆฌ'],
|
144 |
-
'brand': item['๋ธ๋๋๋ช
'],
|
145 |
-
'name': item['์ ํ๋ช
'],
|
146 |
-
'price': item['์ ๊ฐ'],
|
147 |
-
'discount': item['ํ ์ธ์จ'],
|
148 |
-
'image_url': image_url
|
149 |
-
})
|
150 |
-
|
151 |
-
return np.vstack(database_embeddings), database_info
|
152 |
-
|
153 |
-
def find_similar_images(query_embedding, top_k=5):
|
154 |
-
similarities = np.dot(database_embeddings, query_embedding.T).squeeze()
|
155 |
-
top_indices = np.argsort(similarities)[::-1][:top_k]
|
156 |
-
|
157 |
-
results = []
|
158 |
-
for idx in top_indices:
|
159 |
-
results.append({
|
160 |
-
'info': database_info[idx],
|
161 |
-
'similarity': similarities[idx]
|
162 |
-
})
|
163 |
-
|
164 |
-
return results
|
165 |
-
|
166 |
uploaded_file = st.file_uploader("Choose an image...", type="jpg")
|
167 |
if uploaded_file is not None:
|
168 |
image = Image.open(uploaded_file)
|
@@ -195,4 +120,101 @@ if api_key:
|
|
195 |
st.write(f"Discount: {img['info']['discount']}%")
|
196 |
st.write(f"Similarity: {img['similarity']:.2f}")
|
197 |
else:
|
198 |
-
st.warning("Please enter your Roboflow API Key to use the app.")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
12 |
import matplotlib.pyplot as plt
|
13 |
import base64
|
14 |
|
|
|
|
|
|
|
|
|
15 |
# Load model and tokenizer
|
16 |
@st.cache_resource
|
17 |
def load_model():
|
|
|
22 |
|
23 |
model, preprocess_val, tokenizer, device = load_model()
|
24 |
|
25 |
+
# Load and process data
|
26 |
+
@st.cache_data
|
27 |
+
def load_data():
|
28 |
+
with open('musinsa-final.json', 'r', encoding='utf-8') as f:
|
29 |
+
return json.load(f)
|
30 |
+
|
31 |
+
data = load_data()
|
32 |
+
|
33 |
+
# Process database with segmentation
|
34 |
+
@st.cache_data
|
35 |
+
def process_database():
|
36 |
+
database_embeddings = []
|
37 |
+
database_info = []
|
38 |
+
for item in data:
|
39 |
+
image_url = item['์ด๋ฏธ์ง ๋งํฌ'][0]
|
40 |
+
product_id = item.get('\ufeff์ํ ID') or item.get('์ํ ID')
|
41 |
+
|
42 |
+
image = download_and_process_image(image_url)
|
43 |
+
if image is None:
|
44 |
+
continue
|
45 |
+
|
46 |
+
# Save the image temporarily
|
47 |
+
temp_path = f"temp_{product_id}.jpg"
|
48 |
+
image.save(temp_path, 'JPEG')
|
49 |
+
|
50 |
+
segmented_image = segment_image(temp_path)
|
51 |
+
embedding = get_image_embedding(segmented_image)
|
52 |
+
|
53 |
+
database_embeddings.append(embedding)
|
54 |
+
database_info.append({
|
55 |
+
'id': product_id,
|
56 |
+
'category': item['์นดํ
๊ณ ๋ฆฌ'],
|
57 |
+
'brand': item['๋ธ๋๋๋ช
'],
|
58 |
+
'name': item['์ ํ๋ช
'],
|
59 |
+
'price': item['์ ๊ฐ'],
|
60 |
+
'discount': item['ํ ์ธ์จ'],
|
61 |
+
'image_url': image_url
|
62 |
+
})
|
63 |
+
|
64 |
+
return np.vstack(database_embeddings), database_info
|
65 |
+
|
66 |
+
# Initialize database_embeddings and database_info
|
67 |
+
database_embeddings, database_info = process_database()
|
68 |
+
|
69 |
+
def find_similar_images(query_embedding, top_k=5):
|
70 |
+
similarities = np.dot(database_embeddings, query_embedding.T).squeeze()
|
71 |
+
top_indices = np.argsort(similarities)[::-1][:top_k]
|
72 |
+
|
73 |
+
results = []
|
74 |
+
for idx in top_indices:
|
75 |
+
results.append({
|
76 |
+
'info': database_info[idx],
|
77 |
+
'similarity': similarities[idx]
|
78 |
+
})
|
79 |
+
|
80 |
+
return results
|
81 |
|
82 |
# Streamlit app
|
83 |
st.title("Fashion Search App with Segmentation")
|
|
|
88 |
if api_key:
|
89 |
CLIENT = setup_roboflow_client(api_key)
|
90 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
91 |
uploaded_file = st.file_uploader("Choose an image...", type="jpg")
|
92 |
if uploaded_file is not None:
|
93 |
image = Image.open(uploaded_file)
|
|
|
120 |
st.write(f"Discount: {img['info']['discount']}%")
|
121 |
st.write(f"Similarity: {img['similarity']:.2f}")
|
122 |
else:
|
123 |
+
st.warning("Please enter your Roboflow API Key to use the app.")
|
124 |
+
|
125 |
+
|
126 |
+
# Process database with segmentation
|
127 |
+
@st.cache_data
|
128 |
+
def download_and_process_image(image_url):
|
129 |
+
try:
|
130 |
+
response = requests.get(image_url)
|
131 |
+
response.raise_for_status() # Raises an HTTPError for bad responses
|
132 |
+
image = Image.open(BytesIO(response.content))
|
133 |
+
|
134 |
+
# Convert image to RGB mode if it's in RGBA mode
|
135 |
+
if image.mode == 'RGBA':
|
136 |
+
image = image.convert('RGB')
|
137 |
+
|
138 |
+
return image
|
139 |
+
except requests.RequestException as e:
|
140 |
+
st.error(f"Error downloading image: {e}")
|
141 |
+
return None
|
142 |
+
except Exception as e:
|
143 |
+
st.error(f"Error processing image: {e}")
|
144 |
+
return None
|
145 |
+
|
146 |
+
def process_database():
|
147 |
+
database_embeddings = []
|
148 |
+
database_info = []
|
149 |
+
for item in data:
|
150 |
+
image_url = item['์ด๋ฏธ์ง ๋งํฌ'][0]
|
151 |
+
product_id = item.get('\ufeff์ํ ID') or item.get('์ํ ID')
|
152 |
+
|
153 |
+
image = download_and_process_image(image_url)
|
154 |
+
if image is None:
|
155 |
+
continue
|
156 |
+
|
157 |
+
# Save the image temporarily
|
158 |
+
temp_path = f"temp_{product_id}.jpg"
|
159 |
+
image.save(temp_path, 'JPEG')
|
160 |
+
|
161 |
+
segmented_image = segment_image(temp_path)
|
162 |
+
embedding = get_image_embedding(segmented_image)
|
163 |
+
|
164 |
+
database_embeddings.append(embedding)
|
165 |
+
database_info.append({
|
166 |
+
'id': product_id,
|
167 |
+
'category': item['์นดํ
๊ณ ๋ฆฌ'],
|
168 |
+
'brand': item['๋ธ๋๋๋ช
'],
|
169 |
+
'name': item['์ ํ๋ช
'],
|
170 |
+
'price': item['์ ๊ฐ'],
|
171 |
+
'discount': item['ํ ์ธ์จ'],
|
172 |
+
'image_url': image_url
|
173 |
+
})
|
174 |
+
|
175 |
+
return np.vstack(database_embeddings), database_info
|
176 |
+
|
177 |
+
def find_similar_images(query_embedding, top_k=5):
|
178 |
+
similarities = np.dot(database_embeddings, query_embedding.T).squeeze()
|
179 |
+
top_indices = np.argsort(similarities)[::-1][:top_k]
|
180 |
+
|
181 |
+
results = []
|
182 |
+
for idx in top_indices:
|
183 |
+
results.append({
|
184 |
+
'info': database_info[idx],
|
185 |
+
'similarity': similarities[idx]
|
186 |
+
})
|
187 |
+
|
188 |
+
return results
|
189 |
+
|
190 |
+
uploaded_file = st.file_uploader("Choose an image...", type="jpg")
|
191 |
+
if uploaded_file is not None:
|
192 |
+
image = Image.open(uploaded_file)
|
193 |
+
st.image(image, caption='Uploaded Image', use_column_width=True)
|
194 |
+
|
195 |
+
if st.button('Find Similar Items'):
|
196 |
+
with st.spinner('Processing...'):
|
197 |
+
# Save uploaded image temporarily
|
198 |
+
temp_path = "temp_upload.jpg"
|
199 |
+
image.save(temp_path)
|
200 |
+
|
201 |
+
# Segment the uploaded image
|
202 |
+
segmented_image = segment_image(temp_path)
|
203 |
+
st.image(segmented_image, caption='Segmented Image', use_column_width=True)
|
204 |
+
|
205 |
+
# Get embedding for segmented image
|
206 |
+
query_embedding = get_image_embedding(segmented_image)
|
207 |
+
similar_images = find_similar_images(query_embedding)
|
208 |
+
|
209 |
+
st.subheader("Similar Items:")
|
210 |
+
for img in similar_images:
|
211 |
+
col1, col2 = st.columns(2)
|
212 |
+
with col1:
|
213 |
+
st.image(img['info']['image_url'], use_column_width=True)
|
214 |
+
with col2:
|
215 |
+
st.write(f"Name: {img['info']['name']}")
|
216 |
+
st.write(f"Brand: {img['info']['brand']}")
|
217 |
+
st.write(f"Category: {img['info']['category']}")
|
218 |
+
st.write(f"Price: {img['info']['price']}")
|
219 |
+
st.write(f"Discount: {img['info']['discount']}%")
|
220 |
+
st.write(f"Similarity: {img['similarity']:.2f}")
|