Update app.py
Browse files
app.py
CHANGED
@@ -1,3 +1,5 @@
|
|
|
|
|
|
1 |
import streamlit as st
|
2 |
import torch
|
3 |
import bitsandbytes
|
@@ -10,151 +12,159 @@ from my_model.object_detection import detect_and_draw_objects
|
|
10 |
from my_model.captioner.image_captioning import get_caption
|
11 |
from my_model.gen_utilities import free_gpu_resources
|
12 |
from my_model.KBVQA import KBVQA, prepare_kbvqa_model
|
13 |
-
import my_model.utilities.st_config as st_config
|
14 |
|
15 |
|
16 |
|
|
|
17 |
|
|
|
|
|
18 |
|
19 |
-
def
|
20 |
-
|
21 |
-
caption = model.get_caption(img)
|
22 |
-
image_with_boxes, detected_objects_str = model.detect_objects(img)
|
23 |
-
if show_processed_image:
|
24 |
-
st.image(image_with_boxes, use_column_width=True)
|
25 |
-
return caption, detected_objects_str
|
26 |
|
|
|
|
|
27 |
|
|
|
|
|
|
|
|
|
28 |
|
29 |
-
class QuestionAnswering:
|
30 |
-
@staticmethod
|
31 |
-
def answer_question(image, question, caption, detected_objects_str, model):
|
32 |
-
answer = model.generate_answer(question, caption, detected_objects_str)
|
33 |
-
st.image(image, use_column_width=True)
|
34 |
-
st.write(caption)
|
35 |
-
st.write("----------------")
|
36 |
-
st.write(detected_objects_str)
|
37 |
-
return answer
|
38 |
|
39 |
|
40 |
-
def
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
-
|
45 |
-
|
46 |
-
|
47 |
-
|
48 |
-
return True
|
49 |
-
return False
|
50 |
-
|
51 |
-
def set_model_confidence(detection_model):
|
52 |
-
"""Set the confidence level for the detection model."""
|
53 |
-
default_confidence = 0.2 if detection_model == "yolov5" else 0.4
|
54 |
-
confidence_level = st.slider(
|
55 |
-
"Select Detection Confidence Level",
|
56 |
-
min_value=0.1,
|
57 |
-
max_value=0.9,
|
58 |
-
value=default_confidence,
|
59 |
-
step=0.1
|
60 |
-
)
|
61 |
-
st.session_state['kbvqa'].detection_confidence = confidence_level
|
62 |
|
63 |
-
def image_qa_app(
|
64 |
-
|
65 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
66 |
|
|
|
|
|
67 |
cols = st.columns(len(sample_images))
|
68 |
for idx, sample_image_path in enumerate(sample_images):
|
69 |
with cols[idx]:
|
70 |
image = Image.open(sample_image_path)
|
|
|
71 |
if st.button(f'Select Sample Image {idx + 1}', key=f'sample_{idx}'):
|
72 |
-
st.session_state['current_image'] =
|
73 |
st.session_state['qa_history'] = []
|
74 |
st.session_state['analysis_done'] = False
|
75 |
st.session_state['answer_in_progress'] = False
|
76 |
|
|
|
77 |
uploaded_image = st.file_uploader("Or upload an Image", type=["png", "jpg", "jpeg"])
|
78 |
-
st.image(uploaded_image, use_column_width=True)
|
79 |
if uploaded_image is not None:
|
80 |
-
|
|
|
81 |
st.session_state['qa_history'] = []
|
82 |
st.session_state['analysis_done'] = False
|
83 |
st.session_state['answer_in_progress'] = False
|
84 |
|
85 |
-
#
|
86 |
-
if 'current_image'
|
87 |
-
if isinstance(st.session_state['current_image'], str):
|
88 |
-
# If it's a file path from sample images
|
89 |
-
image_to_display = Image.open(st.session_state['current_image'])
|
90 |
-
else:
|
91 |
-
# If it's an uploaded file
|
92 |
-
image_to_display = Image.open(st.session_state['current_image'])
|
93 |
-
st.image(image_to_display, use_column_width=True)
|
94 |
-
else:
|
95 |
-
st.write("No image selected or uploaded.")
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
if st.session_state.get('current_image') and not st.session_state.get('analysis_done', False):
|
101 |
if st.button('Analyze Image'):
|
102 |
-
|
103 |
-
st.session_state['
|
104 |
-
st.session_state['detected_objects_str'] = detected_objects_str
|
105 |
st.session_state['analysis_done'] = True
|
|
|
|
|
|
|
|
|
|
|
106 |
|
107 |
-
|
|
|
108 |
question = st.text_input("Ask a question about this image:")
|
109 |
if st.button('Get Answer'):
|
110 |
-
|
111 |
-
|
112 |
-
question,
|
113 |
-
st.session_state.get('caption', ''),
|
114 |
-
st.session_state.get('detected_objects_str', ''),
|
115 |
-
kbvqa_model
|
116 |
-
)
|
117 |
st.session_state['qa_history'].append((question, answer))
|
118 |
|
119 |
-
|
120 |
-
|
|
|
|
|
|
|
|
|
|
|
121 |
|
122 |
def run_inference():
|
123 |
-
"""Main function to run inference based on the selected method."""
|
124 |
st.title("Run Inference")
|
125 |
|
126 |
method = st.selectbox(
|
127 |
"Choose a method:",
|
128 |
["Fine-Tuned Model", "In-Context Learning (n-shots)"],
|
129 |
-
index=0
|
130 |
)
|
131 |
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
|
136 |
-
|
137 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
138 |
|
139 |
-
if 'kbvqa' not in st.session_state or st.session_state['detection_model'] != detection_model:
|
140 |
-
st.session_state['detection_model'] = detection_model
|
141 |
-
if 'model' not in st.session_state:
|
142 |
-
if st.button('Load Model'):
|
143 |
|
144 |
-
|
145 |
-
|
146 |
-
image_qa_app(st.session_state['kbvqa'])
|
147 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
148 |
|
|
|
|
|
149 |
|
|
|
|
|
150 |
def main():
|
151 |
st.sidebar.title("Navigation")
|
152 |
-
selection = st.sidebar.radio("Go to", ["Home", "Dataset Analysis", "Evaluation Results", "Run Inference", "Dissertation Report"])
|
153 |
|
154 |
if selection == "Home":
|
155 |
-
st.title("MultiModal Learning for
|
156 |
st.write("Home page content goes here...")
|
157 |
-
|
158 |
elif selection == "Dissertation Report":
|
159 |
st.title("Dissertation Report")
|
160 |
st.write("Click the link below to view the PDF.")
|
@@ -166,29 +176,22 @@ def main():
|
|
166 |
mime="application/octet-stream"
|
167 |
)
|
168 |
|
|
|
169 |
elif selection == "Evaluation Results":
|
170 |
st.title("Evaluation Results")
|
171 |
st.write("This is a Place Holder until the contents are uploaded.")
|
172 |
|
|
|
173 |
elif selection == "Dataset Analysis":
|
174 |
st.title("OK-VQA Dataset Analysis")
|
175 |
st.write("This is a Place Holder until the contents are uploaded.")
|
176 |
|
|
|
177 |
elif selection == "Run Inference":
|
178 |
run_inference()
|
179 |
-
|
180 |
-
|
181 |
-
|
182 |
-
|
183 |
-
|
184 |
-
|
185 |
|
186 |
if __name__ == "__main__":
|
187 |
-
main()
|
188 |
-
|
189 |
-
|
190 |
-
|
191 |
-
|
192 |
-
|
193 |
-
|
194 |
-
|
|
|
1 |
+
|
2 |
+
|
3 |
import streamlit as st
|
4 |
import torch
|
5 |
import bitsandbytes
|
|
|
12 |
from my_model.captioner.image_captioning import get_caption
|
13 |
from my_model.gen_utilities import free_gpu_resources
|
14 |
from my_model.KBVQA import KBVQA, prepare_kbvqa_model
|
|
|
15 |
|
16 |
|
17 |
|
18 |
+
def answer_question(image, question, model):
|
19 |
|
20 |
+
answer = model.generate_answer(question, image)
|
21 |
+
return answer
|
22 |
|
23 |
+
def get_caption(image):
|
24 |
+
return "Generated caption for the image"
|
|
|
|
|
|
|
|
|
|
|
25 |
|
26 |
+
def free_gpu_resources():
|
27 |
+
pass
|
28 |
|
29 |
+
# Sample images (assuming these are paths to your sample images)
|
30 |
+
sample_images = ["Files/sample1.jpg", "Files/sample2.jpg", "Files/sample3.jpg",
|
31 |
+
"Files/sample4.jpg", "Files/sample5.jpg", "Files/sample6.jpg",
|
32 |
+
"Files/sample7.jpg"]
|
33 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
34 |
|
35 |
|
36 |
+
def analyze_image(image, model):
|
37 |
+
# Placeholder for your analysis function
|
38 |
+
# This function should prepare captions, detect objects, etc.
|
39 |
+
# For example:
|
40 |
+
# caption = model.get_caption(image)
|
41 |
+
# detected_objects = model.detect_objects(image)
|
42 |
+
# return caption, detected_objects
|
43 |
+
pass
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
44 |
|
45 |
+
def image_qa_app(kbvqa):
|
46 |
+
# Initialize session state for storing the current image and its Q&A history
|
47 |
+
if 'current_image' not in st.session_state:
|
48 |
+
st.session_state['current_image'] = None
|
49 |
+
if 'qa_history' not in st.session_state:
|
50 |
+
st.session_state['qa_history'] = []
|
51 |
+
if 'analysis_done' not in st.session_state:
|
52 |
+
st.session_state['analysis_done'] = False
|
53 |
+
if 'answer_in_progress' not in st.session_state:
|
54 |
+
st.session_state['answer_in_progress'] = False
|
55 |
|
56 |
+
# Display sample images as clickable thumbnails
|
57 |
+
st.write("Choose from sample images:")
|
58 |
cols = st.columns(len(sample_images))
|
59 |
for idx, sample_image_path in enumerate(sample_images):
|
60 |
with cols[idx]:
|
61 |
image = Image.open(sample_image_path)
|
62 |
+
st.image(image, use_column_width=True)
|
63 |
if st.button(f'Select Sample Image {idx + 1}', key=f'sample_{idx}'):
|
64 |
+
st.session_state['current_image'] = image
|
65 |
st.session_state['qa_history'] = []
|
66 |
st.session_state['analysis_done'] = False
|
67 |
st.session_state['answer_in_progress'] = False
|
68 |
|
69 |
+
# Image uploader
|
70 |
uploaded_image = st.file_uploader("Or upload an Image", type=["png", "jpg", "jpeg"])
|
|
|
71 |
if uploaded_image is not None:
|
72 |
+
image = Image.open(uploaded_image)
|
73 |
+
st.session_state['current_image'] = image
|
74 |
st.session_state['qa_history'] = []
|
75 |
st.session_state['analysis_done'] = False
|
76 |
st.session_state['answer_in_progress'] = False
|
77 |
|
78 |
+
# Analyze Image button
|
79 |
+
if st.session_state.get('current_image') and not st.session_state['analysis_done']:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
80 |
if st.button('Analyze Image'):
|
81 |
+
# Perform analysis on the image
|
82 |
+
analyze_image(st.session_state['current_image'], kbvqa)
|
|
|
83 |
st.session_state['analysis_done'] = True
|
84 |
+
st.session_state['processed_image'] = copy.deepcopy(st.session_state['current_image'])
|
85 |
+
|
86 |
+
# Display the current image (unaltered)
|
87 |
+
if st.session_state.get('current_image'):
|
88 |
+
st.image(st.session_state['current_image'], caption='Uploaded Image.', use_column_width=True)
|
89 |
|
90 |
+
# Get Answer button
|
91 |
+
if st.session_state['analysis_done'] and not st.session_state['answer_in_progress']:
|
92 |
question = st.text_input("Ask a question about this image:")
|
93 |
if st.button('Get Answer'):
|
94 |
+
st.session_state['answer_in_progress'] = True
|
95 |
+
answer = answer_question(st.session_state['processed_image'], question, model=kbvqa)
|
|
|
|
|
|
|
|
|
|
|
96 |
st.session_state['qa_history'].append((question, answer))
|
97 |
|
98 |
+
# Display all Q&A
|
99 |
+
for q, a in st.session_state['qa_history']:
|
100 |
+
st.text(f"Q: {q}\nA: {a}\n")
|
101 |
+
|
102 |
+
# Reset the answer_in_progress flag after displaying the answer
|
103 |
+
if st.session_state['answer_in_progress']:
|
104 |
+
st.session_state['answer_in_progress'] = False
|
105 |
|
106 |
def run_inference():
|
|
|
107 |
st.title("Run Inference")
|
108 |
|
109 |
method = st.selectbox(
|
110 |
"Choose a method:",
|
111 |
["Fine-Tuned Model", "In-Context Learning (n-shots)"],
|
112 |
+
index=0 # Default to the first option
|
113 |
)
|
114 |
|
115 |
+
detection_model = st.selectbox(
|
116 |
+
"Choose a model for object detection:",
|
117 |
+
["yolov5", "detic"],
|
118 |
+
index=0 # Default to the first option
|
119 |
+
)
|
120 |
+
|
121 |
+
# Set default confidence based on the selected model
|
122 |
+
default_confidence = 0.2 if detection_model == "yolov5" else 0.4
|
123 |
+
|
124 |
+
# Slider for confidence level
|
125 |
+
confidence_level = st.slider(
|
126 |
+
"Select Detection Confidence Level",
|
127 |
+
min_value=0.1,
|
128 |
+
max_value=0.9,
|
129 |
+
value=default_confidence,
|
130 |
+
step=0.1
|
131 |
+
)
|
132 |
|
|
|
|
|
|
|
|
|
133 |
|
134 |
+
|
135 |
+
# Initialize session state for the model
|
|
|
136 |
|
137 |
+
if method == "Fine-Tuned Model":
|
138 |
+
if 'kbvqa' not in st.session_state:
|
139 |
+
st.session_state['kbvqa'] = None
|
140 |
+
|
141 |
+
# Button to load KBVQA models
|
142 |
+
if st.button('Load KBVQA Model'):
|
143 |
+
if st.session_state['kbvqa'] is not None:
|
144 |
+
st.write("Model already loaded.")
|
145 |
+
else:
|
146 |
+
# Call the function to load models and show progress
|
147 |
+
st.session_state['kbvqa'] = prepare_kbvqa_model(detection_model)
|
148 |
+
|
149 |
+
if st.session_state['kbvqa']:
|
150 |
+
st.write("Model is ready for inference.")
|
151 |
+
|
152 |
+
if st.session_state['kbvqa']:
|
153 |
+
image_qa_app(st.session_state['kbvqa'])
|
154 |
|
155 |
+
else:
|
156 |
+
st.write('Model is not ready for inference yet')
|
157 |
|
158 |
+
|
159 |
+
# Main function
|
160 |
def main():
|
161 |
st.sidebar.title("Navigation")
|
162 |
+
selection = st.sidebar.radio("Go to", ["Home", "Dataset Analysis", "Evaluation Results", "Run Inference", "Dissertation Report", "Object Detection"])
|
163 |
|
164 |
if selection == "Home":
|
165 |
+
st.title("MultiModal Learning for Knowledg-Based Visual Question Answering")
|
166 |
st.write("Home page content goes here...")
|
167 |
+
|
168 |
elif selection == "Dissertation Report":
|
169 |
st.title("Dissertation Report")
|
170 |
st.write("Click the link below to view the PDF.")
|
|
|
176 |
mime="application/octet-stream"
|
177 |
)
|
178 |
|
179 |
+
|
180 |
elif selection == "Evaluation Results":
|
181 |
st.title("Evaluation Results")
|
182 |
st.write("This is a Place Holder until the contents are uploaded.")
|
183 |
|
184 |
+
|
185 |
elif selection == "Dataset Analysis":
|
186 |
st.title("OK-VQA Dataset Analysis")
|
187 |
st.write("This is a Place Holder until the contents are uploaded.")
|
188 |
|
189 |
+
|
190 |
elif selection == "Run Inference":
|
191 |
run_inference()
|
192 |
+
|
193 |
+
elif selection == "Object Detection":
|
194 |
+
run_object_detection()
|
|
|
|
|
|
|
195 |
|
196 |
if __name__ == "__main__":
|
197 |
+
main()
|
|
|
|
|
|
|
|
|
|
|
|
|
|