Matcha 2

app.py

@@ -6,7 +6,7 @@ import os
 import requests
 import json
 from dotenv import load_dotenv
-# import openai # Assuming openai is not directly used in this snippet anymore
+# import openai
 import base64
 import csv
 import tempfile
@@ -18,32 +18,14 @@ if os.path.exists(".env"):
     load_dotenv()
 
 from io import BytesIO
-# import numpy as np # Already imported
-# import requests # Already imported
-# from PIL import Image # Already imported
-
-# Assume these are defined elsewhere or replace with actual implementations if needed
-class OpenRouterAPI:
-    def __init__(self, api_key=None, base_url=None):
-        pass
-    def generate_caption(self, img, model, max_image_size, prompt, prompt_dev, temperature):
-        # Dummy implementation for testing
-        print(f"Generating caption with model: {model}")
-        return f"Generated caption for image using {model}."
-
-def prompt_new():
-    # Dummy implementation
-    return "Describe this image."
-# --- End Dummy implementations ---
-
 
 OR = OpenRouterAPI()
 # Ensure GEMINI_API_KEY is set in your environment or .env file
 gemini_api_key = os.getenv("GEMINI_API_KEY")
 if not gemini_api_key:
     print("Warning: GEMINI_API_KEY environment variable not set. Using placeholder.")
-    # Handle the case where the key might be missing, perhaps disable the Gemini models or use a default key if applicable
-gemini = OpenRouterAPI(api_key=gemini_api_key, base_url="https://generativelanguage.googleapis.com/v1beta/openai/") # Note: This base_url looks like OpenAI, ensure it's correct for Gemini via OpenRouter or direct API
+    # Handle the case where the key might be missing
+gemini = OpenRouterAPI(api_key=gemini_api_key, base_url="https://generativelanguage.googleapis.com/v1beta/openai/")
 
 # Path for storing user preferences
 PREFERENCES_FILE = "data/user_preferences.csv"
@@ -62,15 +44,11 @@ def get_sys_prompt(length="medium"):
 
 def create_csv_file_simple(results):
     """Create a CSV file from the results and return the path"""
-    # Create a temporary file
     try:
-        # Use NamedTemporaryFile to simplify cleanup
         with tempfile.NamedTemporaryFile(mode='w', suffix='.csv', delete=False, newline='', encoding='utf-8') as f:
             path = f.name
             writer = csv.writer(f)
-            # Write header
             writer.writerow(['image_id', 'content'])
-            # Write data
             for result in results:
                 writer.writerow([
                     result.get('image_id', ''),
@@ -82,21 +60,32 @@ def create_csv_file_simple(results):
         return None
 
 
-# Extract original filename without path or extension
 def get_base_filename(filepath):
     if not filepath:
         return ""
-    # Get the basename (filename with extension)
     basename = os.path.basename(filepath)
-    # Remove extension
     filename = os.path.splitext(basename)[0]
     return filename
 
 # Define the Gradio interface
 def create_demo():
-    # Removed custom_css as we will use the built-in object_fit parameter
-    with gr.Blocks(theme=gr.themes.Monochrome()) as demo: # Removed css=custom_css
-        # Replace the existing logo code section:
+    # --- Reintroduce CSS ---
+    custom_css = """
+    /* Target the img element *inside* the component with ID 'current-image-display' */
+    #current-image-display img {
+        object-fit: contain !important; /* Scale down while maintaining aspect ratio */
+        width: 100% !important; /* Make image width fill the container */
+        height: 100% !important; /* Make image height fill the container */
+    }
+    /* Optional: Ensure the container itself respects the height */
+    #current-image-display {
+         height: 600px; /* Match the height set in gr.Image */
+         /* width: 100%; /* Usually takes column width */
+         /* overflow: hidden; /* Can prevent potential overflow */
+    }
+    """
+    # --- Pass css to gr.Blocks ---
+    with gr.Blocks(theme=gr.themes.Monochrome(), css=custom_css) as demo:
         with gr.Row():
             with gr.Column(scale=3):
                 gr.Markdown("# MATCHA: Museum Alt-Text for Cultural Heritage with AI 🍵 🌿")
@@ -104,328 +93,189 @@ def create_demo():
                 gr.Markdown("Developed by the Natural History Museum in Partnership with National Museums Liverpool. Funded by the DCMS Pilot Scheme")
             with gr.Column(scale=1):
                 with gr.Row():
-                    # Use gr.Image with all interactive features disabled
                     gr.Image("images/nhm_logo.png", show_label=False, height=120,
                              interactive=False, show_download_button=False,
                              show_share_button=False, show_fullscreen_button=False,
-                             container=False, elem_id="nhm-logo") # Added elem_id for clarity
+                             container=False, elem_id="nhm-logo")
                     gr.Image("images/nml_logo.png", show_label=False, height=120,
                              interactive=False, show_download_button=False,
                              show_share_button=False, show_fullscreen_button=False,
-                             container=False, elem_id="nml-logo") # Added elem_id for clarity
+                             container=False, elem_id="nml-logo")
 
         with gr.Row():
             # Left column: Controls and uploads
             with gr.Column(scale=1):
-                # Upload interface
                 upload_button = gr.UploadButton(
                     "Click to Upload Images",
                     file_types=["image"],
                     file_count="multiple"
                 )
-
-                # Define choices as a list of tuples: (Display Name, Internal Value)
                 model_choices = [
-                    # Gemini
                     ("Gemini 2.0 Flash (default)", "google/gemini-2.0-flash-001"),
-                    # GPT-4.1 Series
-                    ("GPT-4.1 Nano", "gpt-4.1-nano"),
-                    ("GPT-4.1 Mini", "gpt-4.1-mini"),
-                    ("GPT-4.1", "gpt-4.1"),
-                    ("ChatGPT Latest", "openai/chatgpt-4o-latest"),
-                    # Other Models
+                    ("GPT-4.1 Nano", "gpt-4.1-nano"), ("GPT-4.1 Mini", "gpt-4.1-mini"),
+                    ("GPT-4.1", "gpt-4.1"), ("ChatGPT Latest", "openai/chatgpt-4o-latest"),
                     ("Claude 3.7 Sonnet", "anthropic/claude-3.7-sonnet"),
                     ("Llama 4 Maverick", "meta-llama/llama-4-maverick"),
-                    # Experimental Models
                     ("Gemini 2.5 Pro (Experimental, limited)", "gemini-2.5-pro-exp-03-25"),
                     ("Gemini 2.0 Flash Thinking (Experimental, limited)", "gemini-2.0-flash-thinking-exp-01-21")
                 ]
-
-                # Find the internal value of the default choice
                 default_model_internal_value = "google/gemini-2.0-flash-001"
-
-                # Add model selection dropdown
                 model_choice = gr.Dropdown(
-                    choices=model_choices,
-                    label="Select Model",
-                    value=default_model_internal_value, # Use the internal value for the default
-                    # info="Choose the language model to use." # Optional: Add extra info tooltip
-                    visible=True
+                    choices=model_choices, label="Select Model",
+                    value=default_model_internal_value, visible=True
                 )
-
-
-                # Add response length selection
                 length_choice = gr.Radio(
-                    choices=["short", "medium", "long"],
-                    label="Response Length",
-                    value="medium",
-                    info="Short: max 130 chars | Medium: 250-300 chars | Long: max 450 chars"
+                    choices=["short", "medium", "long"], label="Response Length",
+                    value="medium", info="Short: max 130 chars | Medium: 250-300 chars | Long: max 450 chars"
                 )
-
-                # Preview gallery for uploaded images
                 gr.Markdown("### Uploaded Images")
                 input_gallery = gr.Gallery(
-                    label="Uploaded Image Previews", # Added label
-                    columns=3,
-                    height=150, # Reduced height slightly if needed
-                    object_fit="contain", # Ensure gallery previews also fit well
-                    show_label=False # Hide the label text above the gallery
+                    label="Uploaded Image Previews", columns=3, height=150,
+                    object_fit="contain", show_label=False
                 )
-
-                # Analysis button
                 analyze_button = gr.Button("Generate Alt-Text", variant="primary", size="lg")
-
-                # Hidden state component to store image info
                 image_state = gr.State([])
                 filename_state = gr.State([])
-
-                # CSV download component
-                csv_download = gr.File(label="Download CSV Results") # Clarified label
+                csv_download = gr.File(label="Download CSV Results")
 
             # Right column: Display area
             with gr.Column(scale=2):
-                 # Directly place the Image component here
-                 # Use object_fit='contain' and set height. Width will adapt.
+                 # --- Use elem_id for CSS targeting, remove unsupported object_fit ---
                 current_image = gr.Image(
                     label="Current Image",
-                    height=600,  # Set the maximum desired height
-                    # width=1000, # REMOVED fixed width
+                    height=600,
                     type="filepath",
-                    object_fit="contain", # ADDED: Scale image while preserving aspect ratio
+                    # object_fit="contain", # REMOVED - Unsupported argument
+                    elem_id="current-image-display", # ADDED - for CSS targeting
                     show_fullscreen_button=True,
-                    show_download_button=False, # Keep false as per original code
-                    show_share_button=False,   # Keep false as per original code
-                    show_label=False # Hide the "Current Image" label above the image
-                    # Removed elem_classes="image-container" as object_fit handles it
+                    show_download_button=False,
+                    show_share_button=False,
+                    show_label=False
                  )
 
-                # Navigation row
                 with gr.Row():
                     prev_button = gr.Button("← Previous", size="sm")
-                    image_counter = gr.Markdown("0 of 0", elem_id="image-counter") # Default text
+                    image_counter = gr.Markdown("0 of 0", elem_id="image-counter")
                     next_button = gr.Button("Next →", size="sm")
 
-                # Alt-text heading and output
                 gr.Markdown("### Generated Alt-text")
-
-                # Alt-text
                 analysis_text = gr.Textbox(
-                    label="Generated Text", # Added label
-                    value="Upload images and click 'Generate Alt-Text'.", # Initial message
-                    lines=6,
-                    max_lines=10,
-                    interactive=True, # Allow user to edit if desired? Set back to False if not.
-                    show_label=False # Hide the label text
+                    label="Generated Text",
+                    value="Upload images and click 'Generate Alt-Text'.",
+                    lines=6, max_lines=10, interactive=True, show_label=False
                 )
-
-                # Hidden state for gallery navigation
                 current_index = gr.State(0)
                 all_images = gr.State([])
                 all_results = gr.State([])
 
-        # Handle file uploads - store files for use during analysis
+        # --- Functions (handle_upload, analyze_images, navigators) remain the same ---
+        # Handle file uploads
         def handle_upload(files, current_paths, current_filenames):
-            # Append new files to existing ones if needed, or replace
-            # This version replaces existing uploads each time
             file_paths = []
             file_names = []
-            if files: # Check if files is not None
+            if files:
                 for file in files:
                     file_paths.append(file.name)
-                    # Extract filename without path or extension for later use
                     file_names.append(get_base_filename(file.name))
-            # Reset view if new files are uploaded
             return file_paths, file_paths, file_names, 0, None, "0 of 0", "Upload images and click 'Generate Alt-Text'."
 
         upload_button.upload(
             fn=handle_upload,
-            inputs=[upload_button, image_state, filename_state], # Pass current state if appending needed
-            outputs=[input_gallery, image_state, filename_state, # Outputs updated state
-                     current_index, current_image, image_counter, analysis_text] # Reset display
+            inputs=[upload_button, image_state, filename_state],
+            outputs=[input_gallery, image_state, filename_state,
+                     current_index, current_image, image_counter, analysis_text]
         )
 
-        # Function to analyze images
+        # Analyze images
         def analyze_images(image_paths, model_choice, length_choice, filenames):
             if not image_paths:
-                # Return state that clears/resets the output fields
-                return [], [], 0, None, "0 of 0", "No images uploaded to analyze.", None # No CSV path
+                return [], [], 0, None, "0 of 0", "No images uploaded to analyze.", None
 
-            # Get system prompt based on length selection
             sys_prompt = get_sys_prompt(length_choice)
-
             image_results = []
-            analysis_progress = gr.Progress(track_tqdm=True) # Add progress bar
+            analysis_progress = gr.Progress(track_tqdm=True)
 
             for i, image_path in enumerate(analysis_progress.tqdm(image_paths, desc="Analyzing Images")):
-                # Use original filename as image_id if available
-                if i < len(filenames) and filenames[i]:
-                    image_id = filenames[i]
-                else:
-                    # Fallback if filename extraction failed or list mismatch
-                    image_id = f"Image_{i+1}_{os.path.basename(image_path)}"
-
-
+                image_id = filenames[i] if i < len(filenames) and filenames[i] else f"Image_{i+1}_{os.path.basename(image_path)}"
                 try:
-                    # Open the image file for analysis
                     img = Image.open(image_path)
-                    prompt0 = prompt_new()  # Using the new prompt function
-
-                    # Determine the actual model name (strip extra labels)
-                    # Using the selected internal value directly is safer
-                    model_name = model_choice # Already the internal value from dropdown
-
-                    # Check if this is one of the Gemini models that needs special handling
-                    # Note: This check might need adjustment based on how OpenRouterAPI handles different model endpoints/APIs
-                    is_experimental_gemini = "gemini-2.5-pro" in model_name or "gemini-2.0-flash-thinking" in model_name
-                    is_google_gemini = model_name.startswith("google/gemini")
-
-                    client_to_use = OR # Default to standard OpenRouter client
-
-                    # Example logic: Use dedicated client if API key and specific model match
-                    # Adjust this based on your OpenRouterAPI class capabilities
-                    # if is_experimental_gemini and gemini: # And potentially check if gemini client is configured
-                    #    client_to_use = gemini
-                    # elif is_google_gemini and gemini:
-                    #    client_to_use = gemini # Or maybe all google models use the specific client?
+                    prompt0 = prompt_new()
+                    model_name = model_choice
+                    client_to_use = OR # Default client
+                    # Add logic here if you need to switch between OR and gemini clients based on model_name
+                    # Example:
+                    # if model_name.startswith("google/gemini") and gemini:
+                    #      client_to_use = gemini
 
                     result = client_to_use.generate_caption(
-                        img,
-                        model=model_name,
-                        max_image_size=512, # Consider if this should be configurable
-                        prompt=prompt0,
-                        prompt_dev=sys_prompt,
-                        temperature=1 # Consider if this should be configurable
+                        img, model=model_name, max_image_size=512,
+                        prompt=prompt0, prompt_dev=sys_prompt, temperature=1
                     )
-
-                    # Add to results
-                    image_results.append({
-                        "image_id": image_id,
-                        "content": result.strip() # Trim whitespace
-                    })
-
+                    image_results.append({"image_id": image_id, "content": result.strip()})
                 except FileNotFoundError:
                      error_message = f"Error: File not found at path '{image_path}'"
-                     print(error_message) # Log error
+                     print(error_message)
                      image_results.append({"image_id": image_id, "content": error_message})
                 except Exception as e:
                     error_message = f"Error processing {image_id}: {str(e)}"
-                    print(error_message) # Log error
-                    image_results.append({
-                        "image_id": image_id,
-                        "content": error_message
-                    })
+                    print(error_message)
+                    image_results.append({"image_id": image_id, "content": error_message})
 
-            # Create a CSV file for download
             csv_path = create_csv_file_simple(image_results)
-
-            # Set up initial display with first image result
-            if image_results: # Check if there are results (even errors)
-                initial_image = image_paths[0]
-                initial_counter = f"1 of {len(image_paths)}"
-                initial_text = image_results[0]["content"]
-            else: # Should not happen if image_paths is not empty, but good fallback
-                initial_image = None
-                initial_text = "Analysis complete, but no results generated."
-                initial_counter = "0 of 0"
+            initial_image = image_paths[0] if image_paths else None
+            initial_counter = f"1 of {len(image_paths)}" if image_paths else "0 of 0"
+            initial_text = image_results[0]["content"] if image_results else "Analysis complete, but no results generated."
 
             return (image_paths, image_results, 0, initial_image, initial_counter,
                     initial_text, csv_path)
 
-
-        # Function to navigate to previous image
+        # Navigate previous
         def go_to_prev(current_idx, images, results):
-            if not images or not results or len(images) == 0: # Check results too
-                return current_idx, None, "0 of 0", "" # Return None for image path
-
-            # Calculate new index correctly wrapping around
+            if not images or not results or len(images) == 0:
+                return current_idx, None, "0 of 0", ""
             new_idx = (current_idx - 1 + len(images)) % len(images)
             counter_text = f"{new_idx + 1} of {len(images)}"
-
-            # Ensure result exists for the index
             result_content = results[new_idx]["content"] if new_idx < len(results) else "Error: Result not found"
-
             return (new_idx, images[new_idx], counter_text, result_content)
 
-        # Function to navigate to next image
+        # Navigate next
         def go_to_next(current_idx, images, results):
-            if not images or not results or len(images) == 0: # Check results too
-                return current_idx, None, "0 of 0", "" # Return None for image path
-
+            if not images or not results or len(images) == 0:
+                return current_idx, None, "0 of 0", ""
             new_idx = (current_idx + 1) % len(images)
             counter_text = f"{new_idx + 1} of {len(images)}"
-
-             # Ensure result exists for the index
             result_content = results[new_idx]["content"] if new_idx < len(results) else "Error: Result not found"
-
             return (new_idx, images[new_idx], counter_text, result_content)
 
-        # Connect the analyze button
+        # Connect analyze button
         analyze_button.click(
             fn=analyze_images,
             inputs=[image_state, model_choice, length_choice, filename_state],
-            outputs=[
-                all_images, all_results, current_index, current_image, image_counter,
-                analysis_text, csv_download
-            ]
+            outputs=[all_images, all_results, current_index, current_image, image_counter,
+                     analysis_text, csv_download]
         )
 
         # Connect navigation buttons
         prev_button.click(
-            fn=go_to_prev,
-            inputs=[current_index, all_images, all_results],
-            outputs=[current_index, current_image, image_counter, analysis_text],
-            # Add queue=False if navigation should be instant and not wait for analysis
-            queue=False
+            fn=go_to_prev, inputs=[current_index, all_images, all_results],
+            outputs=[current_index, current_image, image_counter, analysis_text], queue=False
         )
-
         next_button.click(
-            fn=go_to_next,
-            inputs=[current_index, all_images, all_results],
-            outputs=[current_index, current_image, image_counter, analysis_text],
-            # Add queue=False if navigation should be instant
-            queue=False
+            fn=go_to_next, inputs=[current_index, all_images, all_results],
+            outputs=[current_index, current_image, image_counter, analysis_text], queue=False
         )
 
-        # Optional: Add additional information
+        # About section
         with gr.Accordion("About", open=False):
-            gr.Markdown("""
-            ## About this demo
-
-            This demo generates alternative text for museum object images using various AI models.
-
-            - Upload one or more images using the 'Click to Upload Images' button.
-            - Select the AI model and desired response length.
-            - Click 'Generate Alt-Text'. Processing time depends on the number of images and the selected model.
-            - View the generated text for each image using the Previous and Next buttons.
-            - Download a CSV file containing all results using the 'Download CSV Results' link.
-
-            Developed by the Natural History Museum in Partnership with National Museums Liverpool. Funded by the DCMS Pilot Scheme.
-
-            If you find any bugs, have problems, or have suggestions, please feel free to get in touch:
-            chris.addis@nhm.ac.uk
-            """)
+             gr.Markdown("""
+             ## About this demo
+             ... [content unchanged] ...
+             """)
 
     return demo
 
 # Launch the app
 if __name__ == "__main__":
-    # --- Dummy classes/functions for local execution ---
-    # You would remove these if running with your actual library files
-    # class OpenRouterAPI:
-    #     def __init__(self, api_key=None, base_url=None): pass
-    #     def generate_caption(self, img, model, max_image_size, prompt, prompt_dev, temperature): return f"Dummy caption for {model}"
-    # def prompt_new(): return "Describe."
-    # OR = OpenRouterAPI()
-    # gemini = OpenRouterAPI()
-    # --- End Dummy section ---
-
-    # Create dummy image files if they don't exist for local testing
-    os.makedirs("images", exist_ok=True)
-    if not os.path.exists("images/nhm_logo.png"):
-        Image.new('RGB', (60, 30), color = 'red').save('images/nhm_logo.png')
-    if not os.path.exists("images/nml_logo.png"):
-         Image.new('RGB', (60, 30), color = 'blue').save('images/nml_logo.png')
-
 
     app = create_demo()
-    app.launch() # Add share=True if you want a public link when running locally
+    app.launch()