clean-up

main.py

@@ -20,15 +20,15 @@ from utils import preprocess
 img_paths = get_img_paths('image_data','')
 folders = images_by_label(img_paths)
 img_conf = get_embeddings(preprocess,model,device,img_paths,'')
-prompt = 'doctor'
-generic_cat,df = update_plot('CEO', prompt, clip, device, model, folders, img_conf, ['indian businessman','indian businesswoman'])
+
+prompt_x = 'CEO'
+prompt_y =  'doctor'
+generic_cat,df = update_plot(prompt_x, prompt_y, clip, device, model, folders, img_conf, ['indian businessman','indian businesswoman'])
 fig = go.Figure()
-fig = show_scatter(prompt,generic_cat)
+fig = show_scatter(prompt_x, prompt_y,generic_cat)
 
 fig.update_traces(hoverinfo="none", hovertemplate=None)
 
-prompt_x = 'CEO'
-prompt_y =  'doctor'
 def serve_layout():
     session_id = str(uuid.uuid4())
     return update_div(prompt_x,prompt_y,fig,df,img_conf,img_paths,session_id)

utils.py

@@ -10,12 +10,10 @@ from tqdm import tqdm
 import plotly.graph_objects as go
 import pandas as pd
 import base64
-from dash_extensions.enrich import DashProxy, Output, Input, State,no_update, Serverside, html, dcc, \
-    ServersideOutputTransform
+from dash_extensions.enrich import Output, Input, State,Serverside, html, dcc
 import dash
 import shutil
 import json
-import uuid
 
 def download_images(query, number_of_images, output_dir, num_threads, timeout=120):
     """
@@ -95,17 +93,6 @@ def images_by_label(img_paths):
         folders[folder_name].append(img_path)
     return folders
 
-def load_image(img: str):
-    """
-    Load an image and return it.
-    """
-    # if isinstance(img, str):
-    #     img = Image.open(img).convert('RGB')
-    # else:
-    #     img = Image.open(io.BytesIO(img)).convert('RGB')
-    transformed_img = transform_image(img)[:3].unsqueeze(0)
-    return transformed_img
-
 def compute_embedding(preprocess,model,device,img):
     """
     Compute the embedding for an image and return it
@@ -202,31 +189,9 @@ def update_plot(prompt_x, prompt_y, clip, device, model, folders, img_conf, quer
     df = pd.DataFrame.from_dict(new_dict)
     
     return generic_cat, df
-    
-def show_histo(generic_cat):
-    """
-    Show a histogram of similarities for male, female, and generic categories.
-    Args:
-        generic_cat (dict): A dictionary containing similarities for generic categories.
-    Returns:
-        None
-    """
-    # Create a new figure
-    fig = go.Figure()
-    
-    # Add histograms for generic categories
-    for cat, values in generic_cat.items():
-        fig.add_trace(go.Histogram(x=[v[0] for v in values], name=f'{cat}', histnorm='probability'))
-    
-    # Overlay histograms
-    fig.update_layout(barmode='overlay')
-    # Reduce opacity to see both histograms
-    fig.update_traces(opacity=0.55)
-    
-    # Show the figure
-    fig.show()
 
-def show_scatter(prompt, generic_cat):
+
+def show_scatter(prompt_x, prompt_y,generic_cat):
     """
     Show a scatter plot of similarities for male, female, and generic categories.
     Args:
@@ -240,10 +205,8 @@ def show_scatter(prompt, generic_cat):
     fig = go.Figure()
     for cat, values in generic_cat.items():
         fig.add_trace(go.Scatter(x=[v[0] for v in values], y=[v[1] for v in values], mode='markers', name=f'{cat}'))
-    
-    # Update layout
-    fig.update_layout(title="Similarity Scatter Plot", xaxis_title="CEO Similarity Score", yaxis_title=prompt+' similarity score')
-    
+    fig.update_layout(title="Similarity Scatter Plot", xaxis_title=f"{prompt_x} Similarity Score", yaxis_title=f"{prompt_y} Similarity Score")
+    fig.update_traces(hoverinfo="none", hovertemplate=None)
     return fig
 
 def update_div(prompt_x, prompt_y, fig, df, img_conf, img_paths,session_id):
@@ -257,7 +220,6 @@ def update_div(prompt_x, prompt_y, fig, df, img_conf, img_paths,session_id):
         dash_html_components.Div: A Div containing inputs and the graph.
     """
     num_options = [1, 5, 10, 20, 50]
-    print(session_id)
     return html.Div(
         [   
             dcc.Store(id='memory-output', data={'df': df.to_json(),'img_conf':pd.DataFrame.from_dict(img_conf).to_json(),'img_paths':[i for i in img_paths]}),
@@ -306,39 +268,6 @@ def update_div(prompt_x, prompt_y, fig, df, img_conf, img_paths,session_id):
         style={'padding': '20px', 'border': '1px solid #e0e0e0', 'border-radius': '10px'}
     )
 
-
-    
-def display_images_with_preprocessing(img_paths, preprocess, device):
-    """
-    Display original and preprocessed images side by side.
-    Args:
-        img_paths (list): List of paths to image files.
-        preprocess (callable): Function to preprocess the images.
-        device: Device on which the preprocessing will be performed.
-    Returns:
-        None
-    """
-    # Process only the first 5 images
-    for img_path in img_paths:
-        # Preprocess the image
-        x = preprocess(Image.open(img_path)).unsqueeze(0).to(device)
-        # Convert the tensor back to a numpy array
-        preprocessed_image_np = x.squeeze(0).cpu().numpy().transpose((1, 2, 0))
-
-        # Plot the original and preprocessed images side by side
-        fig, axes = plt.subplots(1, 2, figsize=(10, 5))
-        axes[0].imshow(Image.open(img_path))
-        axes[0].set_title('Original Image')
-        axes[0].axis('off')
-
-        axes[1].imshow(preprocessed_image_np)
-        axes[1].set_title('Preprocessed Image')
-        axes[1].axis('off')
-
-        plt.show()
-def download_images_parallel(args):
-    query, n_images_per_class, data_path, num_threads = args
-    download_images(query, n_images_per_class, data_path,num_threads)
 def clean_up(path,session_id):
     data_path = os.path.join(path,session_id)
     if os.path.exists(data_path):
@@ -432,13 +361,7 @@ def register_callbacks(app, df, clip, device, model, folders, img_conf, queries)
                 img_conf = {k:[v['0'],v['1']] for k,v in img_conf.items()}
             folders = images_by_label(img_paths)
             generic_cat, df = update_plot(prompt_x, prompt_y, clip, device, model, folders, img_conf, [query1, query2])
-            
-            fig = go.Figure()
-            for cat, values in generic_cat.items():
-                fig.add_trace(go.Scatter(x=[v[0] for v in values], y=[v[1] for v in values], mode='markers', name=f'{cat}'))
-            fig.update_layout(title="Similarity Scatter Plot", xaxis_title=prompt_x + " Similarity Score",
-                              yaxis_title=prompt_y + ' Similarity score')
-            fig.update_traces(hoverinfo="none", hovertemplate=None)
+            fig = show_scatter(prompt_x, prompt_y,generic_cat)
             data = {'df':df,'img_conf':img_conf,'img_paths':img_paths}
             return [fig,None,Serverside(data)]
         elif 'download-button' in changed_id:
@@ -447,12 +370,7 @@ def register_callbacks(app, df, clip, device, model, folders, img_conf, queries)
             folders = images_by_label(img_paths)
             img_conf = get_embeddings(preprocess,model,device,img_paths,session['session_id'])
             generic_cat, df1 = update_plot(prompt_x, prompt_y, clip, device, model, folders, img_conf, [query1, query2])
-            fig = go.Figure()
-            for cat, values in generic_cat.items():
-                fig.add_trace(go.Scatter(x=[v[0] for v in values], y=[v[1] for v in values], mode='markers', name=f'{cat}'))
-            fig.update_layout(title="Similarity Scatter Plot", xaxis_title=prompt_x + " Similarity Score",
-                              yaxis_title=prompt_y + ' Similarity score')
-            fig.update_traces(hoverinfo="none", hovertemplate=None)
+            fig = show_scatter(prompt_x, prompt_y,generic_cat)
             clean_up('image_data',session['session_id'])
             data = {'df':df1,'img_conf':img_conf,'img_paths':img_paths}
             return [fig,None,Serverside(data)]