Merge branch 'main' of https://huggingface.co/spaces/ludusc/latent-space-theories into main

Home.py

@@ -6,24 +6,28 @@ st.set_page_config(layout='wide')
 
 st.title('About')
 
+st.subheader('General aim of the Ph.D. (to be updated)')
 # INTRO
-intro_text = """This project investigates the nature and nurture of latent spaces, with the aim of formulating a theory of this particular vectorial space. It draws together reflections on the inherent constraints of latent spaces in particular architectures and considers the learning-specific features that emerge.
+intro_text = """
+This project investigates the nature and nurture of latent spaces, with the aim of formulating a theory of this particular vectorial space. It draws together reflections on the inherent constraints of latent spaces in particular architectures and considers the learning-specific features that emerge.
 The thesis concentrates mostly on the second part, exploring different avenues for understanding the space. Using a multitude of vision generative models, it discusses possibilities for the systematic exploration of space, including disentanglement properties and coverage of various guidance methods.
 It also explores the possibility of comparison across latent spaces and investigates the differences and commonalities across different learning experiments. Furthermore, the thesis investigates the role of stochasticity in newer models.
 As a case study, this thesis adopts art historical data, spanning classic art, photography, and modern and contemporary art.
-
-The project aims to interpret the StyleGAN2 model by several techniques. 
-> “What concepts are disentangled in the latent space of StyleGAN2”\n
-> “Can we quantify the complexity of such concepts?”.
-
 """
 st.write(intro_text)
+st.subheader('On this experiment')
+st.write(
+"""The project aims to interpret the StyleGAN3 model trained on Textiles using disentanglement methods. 
+> “What features are disentangled in the latent space of StyleGAN3”\n
+> “Can we quantify the complexity, quality and relations of such features?”.
+""")
 
 # 4 PAGES
 st.subheader('Pages')
-sections_text = """Overall, there are x features in this web app:
-1) Disentanglement visualizer
-2) Concept vectors comparison and aggregation
+sections_text = """Overall, there are 3 features in this web app:
+1) Textiles manipulation 
+2) Features comparison 
+3) Vectors algebra manipulation
 ...
 """
 st.write(sections_text)

backend/disentangle_concepts.py

@@ -7,7 +7,7 @@ from PIL import Image
 
 
 
-def generate_composite_images(model, z, decision_boundaries, lambdas, latent_space='W'):
+def generate_composite_images(model, z, decision_boundaries, lambdas, latent_space='W', negative_colors=None):
     """
     The regenerate_images function takes a model, z, and decision_boundary as input.  It then
     constructs an inverse rotation/translation matrix and passes it to the generator.  The generator
@@ -33,9 +33,19 @@ def generate_composite_images(model, z, decision_boundaries, lambdas, latent_spa
     repetitions = 16 
     z_0 = z
     
-    for decision_boundary, lmbd in zip(decision_boundaries, lambdas):
-        decision_boundary = torch.from_numpy(decision_boundary.copy()).to(device)
-        z_0 = z_0 + int(lmbd) * decision_boundary
+    if negative_colors:
+        for decision_boundary, lmbd, neg_boundary in zip(decision_boundaries, lambdas, negative_colors):
+            decision_boundary = torch.from_numpy(decision_boundary.copy()).to(device)
+            if neg_boundary != 'None':
+                neg_boundary = torch.from_numpy(neg_boundary.copy()).to(device)
+            
+                z_0 = z_0 + int(lmbd) * (decision_boundary - (neg_boundary.T * decision_boundary) * neg_boundary)
+            else:
+                z_0 = z_0 + int(lmbd) * decision_boundary
+    else:            
+        for decision_boundary, lmbd in zip(decision_boundaries, lambdas):
+            decision_boundary = torch.from_numpy(decision_boundary.copy()).to(device)
+            z_0 = z_0 + int(lmbd) * decision_boundary
         
         
     if latent_space == 'Z':

pages/{1_Textiles_Disentanglement.py → 1_Textiles_Manipulation.py}

@@ -20,10 +20,14 @@ BACKGROUND_COLOR = '#bcd0e7'
 SECONDARY_COLOR = '#bce7db'
 
 
-st.title('Disentanglement studies on the Textile Dataset')
+st.title('Disentanglement on Textile Datasets')
 st.markdown(
     """
-    This is a demo of the Disentanglement studies on the [iMET Textiles Dataset](https://www.metmuseum.org/art/collection/search/85531).
+    This is a demo of the Disentanglement experiment on the [iMET Textiles Dataset](https://www.metmuseum.org/art/collection/search/85531).
+    
+    In this page, the user can adjust the colors of textile images generated by an AI by simply traversing the latent space of the AI. 
+    The colors can be adjusted following the human-intuitive encoding of HSV, adjusting the main Hue of the image with an option of 7 colors + Gray,
+    the saturation (the amount of Gray) and the value of the image (the amount of Black).
     """,
     unsafe_allow_html=False,)
     
@@ -73,9 +77,6 @@ if 'num_factors' not in st.session_state:
 if 'best' not in st.session_state:
     st.session_state.best = True
     
-# def on_change_random_input():
-#     st.session_state.image_id = st.session_state.image_id
-
 # ----------------------------- INPUT ----------------------------------
 st.header('Input')
 input_col_1, input_col_2, input_col_3, input_col_4 = st.columns(4)
@@ -83,8 +84,7 @@ input_col_1, input_col_2, input_col_3, input_col_4 = st.columns(4)
 with input_col_1:
    with st.form('image_form'):
         
-        # image_id = st.number_input('Image ID: ', format='%d', step=1)
-        st.write('**Choose or generate a random image to test the disentanglement**')
+        st.write('**Choose or generate a random base image**')
         chosen_image_id_input = st.empty()
         image_id = chosen_image_id_input.number_input('Image ID:', format='%d', step=1, value=st.session_state.image_id)
         
@@ -103,16 +103,15 @@ with input_col_1:
 with input_col_2:
     with st.form('text_form_1'):
         
-        st.write('**Choose color to vary**')
-        type_col = st.selectbox('Color:', tuple(COLORS_LIST), index=7)
-        colors_button = st.form_submit_button('Choose the defined color')
+        st.write('**Choose hue to vary**')
+        type_col = st.selectbox('Hue:', tuple(COLORS_LIST), index=7)
         
         st.write('**Set range of change**')
         chosen_color_lambda_input = st.empty()
         color_lambda = chosen_color_lambda_input.number_input('Lambda:', min_value=-100, step=1, value=7)
-        color_lambda_button = st.form_submit_button('Choose the defined lambda for color')
+        color_lambda_button = st.form_submit_button('Choose the defined hue and lambda')
           
-        if colors_button or color_lambda_button:
+        if color_lambda_button:
             st.session_state.image_id = image_id
             st.session_state.concept_ids = type_col
             st.session_state.color_lambda = color_lambda
@@ -121,24 +120,28 @@ with input_col_2:
 with input_col_3:
     with st.form('text_form'):
         
-        st.write('**Saturation variation**')
+        st.write('**Choose saturation variation**')
         chosen_saturation_lambda_input = st.empty()
         saturation_lambda = chosen_saturation_lambda_input.number_input('Lambda:', min_value=-100, step=1, key=0, value=0)
-        saturation_lambda_button = st.form_submit_button('Choose the defined lambda for saturation')
         
-        st.write('**Value variation**')
+        st.write('**Choose value variation**')
         chosen_value_lambda_input = st.empty()
         value_lambda = chosen_value_lambda_input.number_input('Lambda:', min_value=-100, step=1, key=1, value=0)
-        value_lambda_button = st.form_submit_button('Choose the defined lambda for salue')
+        value_lambda_button = st.form_submit_button('Choose the defined lambda for value and saturation')
         
-        if saturation_lambda_button or value_lambda_button:
+        if value_lambda_button:
             st.session_state.saturation_lambda = int(saturation_lambda)
             st.session_state.value_lambda = int(value_lambda)
             
 with input_col_4:         
     with st.form('text_form_2'):
-        st.write('Use best options')
+        st.write('Use the best vectors (after hyperparameter tuning)')
         best = st.selectbox('Option:', tuple([True, False]), index=0)
+        sign = True
+        num_factors=10
+        cl_method='LR'
+        regularization=0.1
+        extremes=True
         if st.session_state.best is False:
             st.write('Options for StyleSpace (not available for Saturation and Value)')
             sign = st.selectbox('Sign option:', tuple([True, False]), index=1)
@@ -150,10 +153,6 @@ with input_col_4:
             extremes = st.selectbox('Extremes option:', tuple([True, False]), index=1)
             
         choose_options_button = st.form_submit_button('Choose the defined options')
-        # st.write('**Choose a latent space to disentangle**')
-        # # chosen_text_id_input = st.empty()
-        # # concept_id = chosen_text_id_input.text_input('Concept:', value=st.session_state.concept_id)
-        # space_id = st.selectbox('Space:', tuple(['Z', 'W']))
         if choose_options_button:
             st.session_state.best = best
             if st.session_state.best is False:
@@ -181,7 +180,7 @@ with input_col_4:
 # ---------------------------- SET UP OUTPUT ------------------------------
 epsilon_container = st.empty()
 st.header('Image Manipulation')
-st.subheader('Using selected directions')
+st.write('Using selected vectors to modify the original image...')
 
 header_col_1, header_col_2 = st.columns([1,1])
 output_col_1, output_col_2 = st.columns([1,1])
@@ -196,7 +195,7 @@ output_col_1, output_col_2 = st.columns([1,1])
 
 # ---------------------------- DISPLAY COL 1 ROW 1 ------------------------------
 with header_col_1:
-    st.write(f'Original image')
+    st.write(f'### Original image')
 
 with header_col_2:
     if st.session_state.best:
@@ -212,8 +211,15 @@ with header_col_2:
         tmp_sat = concept_vectors[concept_vectors['color'] == 'Saturation'][concept_vectors['extremes'] == st.session_state.extremes]
         saturation_separation_vector, performance_saturation = tmp_sat.reset_index().loc[0, ['vector', 'score']]
         
-    st.write(f'Change in {st.session_state.concept_ids} of {np.round(st.session_state.color_lambda, 2)}, in saturation of {np.round(st.session_state.saturation_lambda, 2)}, in value of {np.round(st.session_state.value_lambda, 2)}. - Performance color vector: {performance_color}, saturation vector: {performance_saturation/100}, value vector: {performance_value/100}')
-
+    st.write('### Modified image')
+    st.write(f"""
+            Change in hue: {st.session_state.concept_ids} of amount: {np.round(st.session_state.color_lambda, 2)}, 
+            in: saturation of amount: {np.round(st.session_state.saturation_lambda, 2)}, 
+            in: value of amount: {np.round(st.session_state.value_lambda, 2)}.\
+            Verification performance of hue vector: {performance_color}, 
+            saturation vector: {performance_saturation/100},
+            value vector: {performance_value/100}""")
+    
 # ---------------------------- DISPLAY COL 2 ROW 1 ------------------------------
 
 if st.session_state.space_id == 'Z':
@@ -229,3 +235,4 @@ with output_col_1:
 with output_col_2:
     image_updated = generate_composite_images(model, original_image_vec, [color_separation_vector, saturation_separation_vector, value_separation_vector], lambdas=[st.session_state.color_lambda, st.session_state.saturation_lambda, st.session_state.value_lambda])
     st.image(image_updated)
+    

pages/{2_Colours_comparison.py → 2_Network_comparison.py}

@@ -24,7 +24,11 @@ st.set_page_config(layout='wide')
 
 st.title('Comparison among color directions')
 st.write('> **How do the color directions relate to each other?**')
-st.write('> **What is their joint impact on the image?**')
+st.write("""
+         This page provides a simple network-based framework to inspect the vector similarity (cosine similarity) among the found color vectors.
+         The nodes are the colors chosen for comparison and the strength of the edge represents the similarity.
+         
+         """)
 
         
 annotations_file = './data/textile_annotated_files/seeds0000-100000_S.pkl'
@@ -46,10 +50,8 @@ with dnnlib.util.open_url('./data/textile_model_files/network-snapshot-005000.pk
 
 COLORS_LIST = ['Gray', 'Red Orange', 'Yellow', 'Green', 'Light Blue', 'Blue', 'Purple', 'Pink', 'Saturation', 'Value']
 
-if 'image_id' not in st.session_state:
-    st.session_state.image_id = 0
 if 'concept_ids' not in st.session_state:
-    st.session_state.concept_ids = [COLORS_LIST[-1], COLORS_LIST[-2], ]
+    st.session_state.concept_ids = COLORS_LIST
 if 'sign' not in st.session_state:
     st.session_state.sign = False
 if 'extremes' not in st.session_state:
@@ -60,10 +62,8 @@ if 'cl_method' not in st.session_state:
     st.session_state.cl_method = False
 if 'num_factors' not in st.session_state:
     st.session_state.num_factors = False
-
-
-if 'space_id' not in st.session_state:
-    st.session_state.space_id = 'W'
+if 'best' not in st.session_state:
+    st.session_state.best = True
 
 # ----------------------------- INPUT ----------------------------------
 st.header('Input')
@@ -76,7 +76,7 @@ with input_col_1:
         st.write('**Choose a series of colors to compare**')
         # chosen_text_id_input = st.empty()
         # concept_id = chosen_text_id_input.text_input('Concept:', value=st.session_state.concept_id)
-        concept_ids = st.multiselect('Color (including Saturation and Value):', tuple(COLORS_LIST), default=[COLORS_LIST[-1], COLORS_LIST[-2], ])
+        concept_ids = st.multiselect('Color (including Saturation and Value):', tuple(COLORS_LIST), default=COLORS_LIST)
         choose_text_button = st.form_submit_button('Choose the defined colors')
         
         if choose_text_button:
@@ -85,27 +85,33 @@ with input_col_1:
             
 with input_col_2:
     with st.form('text_form_1'):
-        st.write('Options for StyleSpace (not available for Saturation and Value)')
-        sign = st.selectbox('Sign option:', tuple([True, False]), index=1)
-        num_factors = st.selectbox('Number of factors option:', tuple([1, 5, 10, 20, False]), index=4)
-        st.write('Options for InterFaceGAN (not available for Saturation and Value)')
-        cl_method = st.selectbox('Classification method option:', tuple(['LR', 'SVM', False]), index=2)
-        regularization = st.selectbox('Regularization option:', tuple([0.1, 1.0, False]), index=2)
-        st.write('Options for InterFaceGAN (only for Saturation and Value)')
-        extremes = st.selectbox('Extremes option:', tuple([True, False]), index=1)
-        
+        st.write('Use the best vectors (after hyperparameter tuning)')
+        best = st.selectbox('Option:', tuple([True, False]), index=0)
+        sign = True
+        num_factors=10
+        cl_method='LR'
+        regularization=0.1
+        extremes=True
+        if st.session_state.best is False:
+            st.write('Options for StyleSpace (not available for Saturation and Value)')
+            sign = st.selectbox('Sign option:', tuple([True, False]), index=1)
+            num_factors = st.selectbox('Number of factors option:', tuple([1, 5, 10, 20, False]), index=4)
+            st.write('Options for InterFaceGAN (not available for Saturation and Value)')
+            cl_method = st.selectbox('Classification method option:', tuple(['LR', 'SVM', False]), index=2)
+            regularization = st.selectbox('Regularization option:', tuple([0.1, 1.0, False]), index=2)
+            st.write('Options for InterFaceGAN (only for Saturation and Value)')
+            extremes = st.selectbox('Extremes option:', tuple([True, False]), index=1)
+            
         choose_options_button = st.form_submit_button('Choose the defined options')
-        # st.write('**Choose a latent space to disentangle**')
-        # # chosen_text_id_input = st.empty()
-        # # concept_id = chosen_text_id_input.text_input('Concept:', value=st.session_state.concept_id)
-        # space_id = st.selectbox('Space:', tuple(['Z', 'W']))
         if choose_options_button:
-            st.session_state.sign = sign
-            st.session_state.num_factors = num_factors
-            st.session_state.cl_method = cl_method
-            st.session_state.regularization = regularization
-            st.session_state.extremes = extremes
-            
+            st.session_state.best = best
+            if st.session_state.best is False:
+                st.session_state.sign = sign
+                st.session_state.num_factors = num_factors
+                st.session_state.cl_method = cl_method
+                st.session_state.regularization = regularization
+                st.session_state.extremes = extremes
+        
 # ---------------------------- SET UP OUTPUT ------------------------------
 epsilon_container = st.empty()
 st.header('Comparison')
@@ -115,23 +121,28 @@ header_col_1, header_col_2 = st.columns([3,1])
 output_col_1, output_col_2 = st.columns([3,1])
 
 # ---------------------------- DISPLAY COL 1 ROW 1 ------------------------------
-tmp = concept_vectors[concept_vectors['color'].isin(st.session_state.concept_ids)]
-tmp = tmp[tmp['sign'] == st.session_state.sign][tmp['extremes'] == st.session_state.extremes][tmp['num_factors'] == st.session_state.num_factors][tmp['cl_method'] == st.session_state.cl_method][tmp['regularization'] == st.session_state.regularization]
+if st.session_state.best:
+    tmp = concept_vectors[concept_vectors['color'].isin(st.session_state.concept_ids)].groupby('color').first().reset_index()
+else:
+    tmp = concept_vectors[concept_vectors['color'].isin(st.session_state.concept_ids)]
+    tmp = tmp[tmp['sign'] == st.session_state.sign][tmp['extremes'] == st.session_state.extremes][tmp['num_factors'] == st.session_state.num_factors][tmp['cl_method'] == st.session_state.cl_method][tmp['regularization'] == st.session_state.regularization]
+
 info = tmp.loc[:, ['vector', 'score', 'color', 'kwargs']].values
 concept_ids = [i[2] for i in info] #+ ' ' + i[3]
     
 with header_col_1:
-    st.write('Similarity graph')
+    st.write('### Similarity graph')
 
 with header_col_2:
-    st.write('Information')
+    st.write('### Information')
     
 with output_col_2:
     for i,concept_id in enumerate(concept_ids):
-        st.write(f'Color {info[i][2]} - Settings: {info[i][3]} Performance of the color vector: {info[i][1]}')# - Nodes {",".join(list(imp_nodes))}')
+        st.write(f'''Color: {info[i][2]}.\
+                     Settings: {info[i][3]}\
+                     ''')
 
 with output_col_1:
-    
     edges = []
     for i in range(len(concept_ids)):
         for j in range(len(concept_ids)):

pages/3_Vectors_algebra.py

@@ -0,0 +1,163 @@
+import streamlit as st
+import pickle
+import pandas as pd
+import numpy as np
+import random
+import torch
+
+from matplotlib.backends.backend_agg import RendererAgg
+
+from backend.disentangle_concepts import *
+import torch_utils
+import dnnlib
+import legacy
+
+_lock = RendererAgg.lock
+
+
+st.set_page_config(layout='wide')
+BACKGROUND_COLOR = '#bcd0e7'
+SECONDARY_COLOR = '#bce7db'
+
+
+st.title('Vector algebra using disentangled vectors')
+st.markdown(
+    """
+    This page offers the possibility to edit the colors of a given textile image using vector algebra and projections.
+    It allows to select several colors to move towards and against (selecting a positive or negative lambda). 
+    Furthermore, it offers the possibility of conditional manipulation, by moving in the direction of a color n1 without affecting the color n2. 
+    This is done using a projected direction n1 - (n1.T n2) n2. 
+    """,
+    unsafe_allow_html=False,)
+    
+annotations_file = './data/textile_annotated_files/seeds0000-100000_S.pkl'
+with open(annotations_file, 'rb') as f:
+    annotations = pickle.load(f)
+
+concept_vectors = pd.read_csv('./data/stored_vectors/scores_colors_hsv.csv')
+concept_vectors['vector'] = [np.array([float(xx) for xx in x]) for x in concept_vectors['vector'].str.split(', ')]
+concept_vectors['score'] = concept_vectors['score'].astype(float)
+
+concept_vectors = concept_vectors.sort_values('score', ascending=False).reset_index()
+
+with dnnlib.util.open_url('./data/textile_model_files/network-snapshot-005000.pkl') as f:
+    model = legacy.load_network_pkl(f)['G_ema'].to('cpu') # type: ignore
+
+COLORS_LIST = ['Gray', 'Red Orange', 'Yellow', 'Green', 'Light Blue', 'Blue', 'Purple', 'Pink', 'Saturation', 'Value']
+COLORS_NEGATIVE = COLORS_LIST + ['None']
+
+if 'image_id' not in st.session_state:
+    st.session_state.image_id = 52921
+if 'colors' not in st.session_state:
+    st.session_state.colors = [COLORS_LIST[5], COLORS_LIST[7]]
+if 'non_colors' not in st.session_state:
+    st.session_state.non_colors = ['None']
+if 'color_lambda' not in st.session_state:
+    st.session_state.color_lambda = [5]
+
+# ----------------------------- INPUT ----------------------------------
+epsilon_container = st.empty()
+st.header('Image Manipulation with Vector Algebra')
+
+header_col_1, header_col_2, header_col_3, header_col_4 = st.columns([1,1,1,1])
+input_col_1, output_col_2, output_col_3, input_col_4 = st.columns([1,1,1,1])
+
+# --------------------------- INPUT column 1 ---------------------------
+with input_col_1:
+   with st.form('image_form'):
+        
+        # image_id = st.number_input('Image ID: ', format='%d', step=1)
+        st.write('**Choose or generate a random image**')
+        chosen_image_id_input = st.empty()
+        image_id = chosen_image_id_input.number_input('Image ID:', format='%d', step=1, value=st.session_state.image_id)
+        
+        choose_image_button = st.form_submit_button('Choose the defined image')
+        random_id = st.form_submit_button('Generate a random image')
+
+        if random_id:
+            image_id = random.randint(0, 100000)
+            st.session_state.image_id = image_id
+            chosen_image_id_input.number_input('Image ID:', format='%d', step=1, value=st.session_state.image_id)
+            
+        if choose_image_button:
+            image_id = int(image_id)
+            st.session_state.image_id = image_id
+
+with header_col_1:
+    st.write('### Input image selection')
+
+original_image_vec = annotations['w_vectors'][st.session_state.image_id]
+img = generate_original_image(original_image_vec, model)
+
+with output_col_2:
+    st.image(img)
+
+with header_col_2:
+    st.write('### Original image')
+    
+with input_col_4:
+    with st.form('text_form_1'):
+        
+        st.write('**Colors to vary (including Saturation and Value)**')
+        colors = st.multiselect('Color:', tuple(COLORS_LIST), default=[COLORS_LIST[5], COLORS_LIST[7]])
+        colors_button = st.form_submit_button('Choose the defined colors')
+        
+        st.session_state.image_id = image_id
+        st.session_state.colors = colors
+        st.session_state.color_lambda = [5]*len(colors)
+        st.session_state.non_colors = ['None']*len(colors)
+            
+        lambdas = []
+        negative_cols = []
+        for color in colors:
+            st.write('### '+color )
+            st.write('**Set range of change (can be negative)**')
+            chosen_color_lambda_input = st.empty()
+            color_lambda = chosen_color_lambda_input.number_input('Lambda:', min_value=-100, step=1, value=5, key=color+'_number')
+            lambdas.append(color_lambda)
+                
+            st.write('**Set dimensions of change to not consider**')
+            chosen_color_negative_input = st.empty()
+            color_negative = chosen_color_negative_input.selectbox('Color:', tuple(COLORS_NEGATIVE), index=len(COLORS_NEGATIVE)-1, key=color+'_noncolor')
+            negative_cols.append(color_negative)
+                
+        lambdas_button = st.form_submit_button('Submit options')
+        if lambdas_button:
+            st.session_state.color_lambda = lambdas
+            st.session_state.non_colors = negative_cols
+                
+        
+with header_col_4:
+    st.write('### Color settings')
+# print(st.session_state.colors)
+# print(st.session_state.color_lambda)
+# print(st.session_state.non_colors)
+
+# ---------------------------- DISPLAY COL 1 ROW 1 ------------------------------
+
+with header_col_3:
+    separation_vectors = []
+    for col in st.session_state.colors:
+        separation_vector, score_1 = concept_vectors[concept_vectors['color'] == col].reset_index().loc[0, ['vector', 'score']]
+        separation_vectors.append(separation_vector)
+    
+    negative_separation_vectors = []
+    for non_col in st.session_state.non_colors:
+        if non_col != 'None':
+            negative_separation_vector, score_2 = concept_vectors[concept_vectors['color'] == non_col].reset_index().loc[0, ['vector', 'score']]
+            negative_separation_vectors.append(negative_separation_vector)
+        else:
+            negative_separation_vectors.append('None')
+    ## n1 − (n1T n2)n2
+    # print(negative_separation_vectors, separation_vectors)
+    st.write('### Output Image')
+    st.write(f'''Change in colors: {str(st.session_state.colors)},\
+             without affecting colors {str(st.session_state.non_colors)}''')
+
+# ---------------------------- DISPLAY COL 2 ROW 1 ------------------------------
+
+with output_col_3:
+    image_updated = generate_composite_images(model, original_image_vec, separation_vectors, 
+                                              lambdas=st.session_state.color_lambda, 
+                                              negative_colors=negative_separation_vectors)
+    st.image(image_updated)