Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,67 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
from PIL import Image
|
3 |
+
impor os
|
4 |
+
import conch
|
5 |
+
from conch.open_clip_custom import create_model_from_pretrained, get_tokenizer, tokenize
|
6 |
+
|
7 |
+
import torch
|
8 |
+
from huggingface_hub import login
|
9 |
+
hf=os.getenv('hf')
|
10 |
+
login(hf)
|
11 |
+
# Load the pretrained model and transforms
|
12 |
+
#model = create_model("hf-hub:MahmoodLab/UNI", pretrained=True, init_values=1e-5, dynamic_img_size=True)
|
13 |
+
# Load the model
|
14 |
+
@st.cache_resource
|
15 |
+
def load_model():
|
16 |
+
model, preprocess = create_model_from_pretrained('conch_ViT-B-16', "hf_hub:MahmoodLab/conch",hf_auth_token=hf)
|
17 |
+
return model, preprocess
|
18 |
+
|
19 |
+
model, preprocess = load_model()
|
20 |
+
|
21 |
+
st.title("CONCH - Image Captioning and Retrieval")
|
22 |
+
|
23 |
+
# Upload an image
|
24 |
+
uploaded_file = st.file_uploader("Upload an image", type=["jpg", "png", "jpeg"])
|
25 |
+
|
26 |
+
if uploaded_file:
|
27 |
+
# Display uploaded image
|
28 |
+
image = Image.open(uploaded_file)
|
29 |
+
st.image(image, caption='Uploaded Image', use_column_width=True)
|
30 |
+
|
31 |
+
# Preprocess and get image embeddings
|
32 |
+
image = preprocess(image).unsqueeze(0)
|
33 |
+
|
34 |
+
with torch.no_grad():
|
35 |
+
image_embs = model.encode_image(image, proj_contrast=True, normalize=True)
|
36 |
+
|
37 |
+
st.write("Image embeddings generated successfully.")
|
38 |
+
|
39 |
+
# Text retrieval example
|
40 |
+
# Number of inputs to display (can be dynamic)
|
41 |
+
num_inputs = st.number_input("How many text inputs?", min_value=1, max_value=10, value=3)
|
42 |
+
|
43 |
+
# List to store user inputs
|
44 |
+
input_list = []
|
45 |
+
|
46 |
+
# Dynamically generate the text input fields
|
47 |
+
for i in range(num_inputs):
|
48 |
+
user_input = st.text_input(f"Input Text {i+1}")
|
49 |
+
input_list.append(user_input)
|
50 |
+
|
51 |
+
populated_status = ["Populated" if text.strip() else "Empty" for text in input_list]
|
52 |
+
if "Populated" in populated_status:
|
53 |
+
# Tokenize the text
|
54 |
+
tokenizer = get_tokenizer() # load tokenizer
|
55 |
+
text_tokens = tokenize(texts=input_list, tokenizer=tokenizer) # tokenize the text
|
56 |
+
text_embs = model.encode_text(text_tokens)
|
57 |
+
|
58 |
+
|
59 |
+
#with torch.no_grad():
|
60 |
+
# text_embs = model.encode_text(tokens, proj_contrast=True, normalize=True)
|
61 |
+
|
62 |
+
st.write("Text embeddings generated successfully.")
|
63 |
+
|
64 |
+
# Perform similarity check
|
65 |
+
similarity = torch.cosine_similarity(image_embs, text_embs)
|
66 |
+
st.write("Similarity check completed.")
|
67 |
+
st.write(similarity)
|