Update app.py

app.py

@@ -1,77 +1,47 @@
 import gradio as gr
-import torch
-from PIL import Image
-import pandas as pd
-from lavis.models import load_model_and_preprocess
-from lavis.processors import load_processor
-from transformers import AutoTokenizer, AutoModelForCausalLM, AutoProcessor
 import tensorflow as tf
 import tensorflow_hub as hub
-import io
-from sklearn.metrics.pairwise import cosine_similarity
-import tempfile
-import logging
-
-# Configure logging
-logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
-
-# Load model and preprocessors for Image-Text Matching (LAVIS)
-device = torch.device("cuda") if torch.cuda.is_available() else "cpu"
-model_itm, vis_processors, text_processors = load_model_and_preprocess("blip2_image_text_matching", "pretrain", device=device, is_eval=True)
-
-# Load tokenizer and model for Image Captioning (TextCaps)
-git_processor_large_textcaps = AutoProcessor.from_pretrained("microsoft/git-large-r-textcaps")
-git_model_large_textcaps = AutoModelForCausalLM.from_pretrained("microsoft/git-large-r-textcaps")
-
-# Load Universal Sentence Encoder model for textual similarity calculation
-embed = hub.load("https://tfhub.dev/google/universal-sentence-encoder/4")
-
-# Define a function to compute textual similarity between caption and statement
-def compute_textual_similarity(caption, statement):
-    # Convert caption and statement into sentence embeddings
-    caption_embedding = embed([caption])[0].numpy()
-    statement_embedding = embed([statement])[0].numpy()
+import numpy as np
+import pandas as pd
+from transformers import GitProcessor, GitModel, GitConfig, ImageFeatureProcessor
+from PIL import Image
 
-    # Calculate cosine similarity between sentence embeddings
-    similarity_score = cosine_similarity([caption_embedding], [statement_embedding])[0][0]
-    return similarity_score
+# Load models and processors
+git_config = GitConfig.from_pretrained("microsoft/git-large-r")
+git_processor_large_textcaps = GitProcessor.from_pretrained("microsoft/git-large-r")
+git_model_large_textcaps = GitModel.from_pretrained("microsoft/git-large-r")
+itm_model = hub.load("https://tfhub.dev/google/LaViT/1")
+use_model = hub.load("https://tfhub.dev/google/universal-sentence-encoder-large/5")
 
 # Read statements from the external file 'statements.txt'
 with open('statements.txt', 'r') as file:
     statements = file.read().splitlines()
 
-# Function to compute ITM scores for the image-statement pair
-def compute_itm_score(image, statement):
-    logging.info('Starting compute_itm_score')
-    pil_image = Image.fromarray(image.astype('uint8'), 'RGB')
-    img = vis_processors["eval"](pil_image.convert("RGB")).unsqueeze(0).to(device)
-    # Pass the statement text directly to model_itm
-    itm_output = model_itm({"image": img, "text_input": statement}, match_head="itm")
-    itm_scores = torch.nn.functional.softmax(itm_output, dim=1)
-    score = itm_scores[:, 1].item()
-    logging.info('Finished compute_itm_score')
-    return score
-
+# Function to generate image caption
 def generate_caption(processor, model, image):
-    logging.info('Starting generate_caption')
-    inputs = processor(images=image, return_tensors="pt").to(device)
-    generated_ids = model.generate(pixel_values=inputs.pixel_values, max_length=50)
-    generated_caption = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
-    logging.info('Finished generate_caption')
-    return generated_caption
+    inputs = processor(images=image, return_tensors="pt")
+    outputs = model(**inputs)
+    caption = processor.batch_decode(outputs.logits.argmax(-1), skip_special_tokens=True)
+    return caption[0]
 
-def save_dataframe_to_csv(df):
-    csv_buffer = io.StringIO()
-    df.to_csv(csv_buffer, index=False)
-    csv_string = csv_buffer.getvalue()
+# Function to compute textual similarity
+def compute_textual_similarity(caption, statement):
+    captions_embeddings = use_model([caption])[0].numpy()
+    statements_embeddings = use_model([statement])[0].numpy()
+    similarity_score = np.inner(captions_embeddings, statements_embeddings)
+    return similarity_score[0]
 
-    # Save the CSV string to a temporary file
-    with tempfile.NamedTemporaryFile(mode="w+", delete=False, suffix=".csv") as temp_file:
-        temp_file.write(csv_string)
-        temp_file_path = temp_file.name # Get the file path
+# Function to compute ITM score
+def compute_itm_score(image, statement):
+    image_features = itm_model(image)
+    statement_features = use_model([statement])[0].numpy()
+    similarity_score = np.inner(image_features, statement_features)
+    return similarity_score[0][0]
 
-    # Return the file path (no need to reopen the file with "rb" mode)
-    return temp_file_path
+# Function to save DataFrame to CSV
+def save_dataframe_to_csv(df):
+    csv_data = df.to_csv(index=False)
+    return csv_data
 
 # Main function to perform image captioning and image-text matching for multiple images
 def process_images_and_statements(files):
@@ -81,7 +51,10 @@ def process_images_and_statements(files):
     if isinstance(files, list):
         files = {f.name: f for f in files}
     
-    for file_name, image in files.items():
+    for file_name, image_file in files.items():
+        # Convert the image file to a PIL image
+        image = Image.open(image_file)
+        
         caption = generate_caption(git_processor_large_textcaps, git_model_large_textcaps, image)
         for statement in statements:
             textual_similarity_score = compute_textual_similarity(caption, statement) * 100
@@ -99,6 +72,7 @@ def process_images_and_statements(files):
     csv_results = save_dataframe_to_csv(results_df)
     return results_df, csv_results
 
+# Gradio interface with File input to receive
 # Gradio interface with File input to receive multiple images and file names
 image_input = gr.inputs.File(file_count="multiple", type="file", label="Upload Images")
 output_df = gr.outputs.Dataframe(type="pandas", label="Results")
@@ -110,7 +84,8 @@ iface = gr.Interface(
     outputs=[output_df, output_csv],
     title="Image Captioning and Image-Text Matching",
     theme='sudeepshouche/minimalist',
-    css=".output { flex-direction: column; } .output .outputs { width: 100%; }" # Custom CSS
+    css=".output { flex-direction: column; } .output .outputs { width: 100%; }", # Custom CSS
+    capture_session=True,  # Capture errors and exceptions in Gradio interface
 )
 
 iface.launch(debug=True)