Update app.py

app.py

@@ -6,7 +6,7 @@ import torchvision.transforms as transforms
 from transformers import VisionEncoderDecoderModel, AutoTokenizer, AutoModelForSeq2SeqLM
 
 # Load the models
-caption_model = VisionEncoderDecoderModel.from_pretrained('Mayada/AIC-transformer')  # Your model on Hugging Face
+caption_model = VisionEncoderDecoderModel.from_pretrained('/content/drive/MyDrive/ICModel')
 caption_tokenizer = AutoTokenizer.from_pretrained('aubmindlab/bert-base-arabertv02')
 question_model = AutoModelForSeq2SeqLM.from_pretrained("Mihakram/AraT5-base-question-generation")
 question_tokenizer = AutoTokenizer.from_pretrained("Mihakram/AraT5-base-question-generation")
@@ -23,10 +23,9 @@ inference_transforms = transforms.Compose([
     normalize
 ])
 
-# Load the dictionary (use it from your Hugging Face Space or include in the repo)
-dictionary = {
-    "caption": "alternative_caption"  # Replace with your actual dictionary
-}
+# Load the dictionary
+with open("/content/drive/MyDrive/DICTIONARY (3).txt", "r", encoding="utf-8") as file:
+    dictionary = dict(line.strip().split("\t") for line in file)
 
 # Function to correct words in the caption using the dictionary
 def correct_caption(caption):
@@ -67,49 +66,67 @@ def generate_questions(context, answer):
         'question: ', ' ') for g in generated_ids]
     return questions
 
-# Gradio Interface Function
+# Define the Gradio interface with Seafoam theme
+class Seafoam(Base):
+    pass
+
+seafoam = Seafoam()
+
 def caption_question_interface(image):
+    # Generate captions
     captions = generate_captions(image)
+
+    # Correct captions using the dictionary
     corrected_captions = [correct_caption(caption) for caption in captions]
+
+    # Generate questions for each caption
     questions_with_answers = []
-    
     for caption in corrected_captions:
         words = caption.split()
+        # Generate questions for the first word
         if len(words) > 0:
             answer = words[0]
             question = generate_questions(caption, answer)
             questions_with_answers.extend([(q, answer) for q in question])
+        # Generate questions for the second word
         if len(words) > 1:
             answer = words[1]
             question = generate_questions(caption, answer)
             questions_with_answers.extend([(q, answer) for q in question])
+        # Generate questions for the second word + first word
         if len(words) > 1:
             answer = " ".join(words[:2])
             question = generate_questions(caption, answer)
             questions_with_answers.extend([(q, answer) for q in question])
+        # Generate questions for the third word
         if len(words) > 2:
             answer = words[2]
             question = generate_questions(caption, answer)
             questions_with_answers.extend([(q, answer) for q in question])
+        # Generate questions for the fourth word
         if len(words) > 3:
             answer = words[3]
             question = generate_questions(caption, answer)
             questions_with_answers.extend([(q, answer) for q in question])
 
+    # Format questions with answers
     formatted_questions = [f"Question: {q}\nAnswer: {a}" for q, a in questions_with_answers]
     formatted_questions = "\n".join(formatted_questions)
 
+    # Return the generated captions and formatted questions with answers
     return "\n".join(corrected_captions), formatted_questions
 
 gr_interface = gr.Interface(
     fn=caption_question_interface,
-    inputs=gr.inputs.Image(type="pil", label="Input Image"),
+    inputs=gr.Image(type="pil", label="Input Image"), 
     outputs=[
-        gr.outputs.Textbox(label="Generated Captions"),
-        gr.outputs.Textbox(label="Generated Questions and Answers")
+        gr.Textbox(label="Generated Captions"), 
+        gr.Textbox(label="Generated Questions and Answers")
     ],
     title="Image Captioning and Question Generation",
-    description="Generate captions and questions for images using pre-trained models."
+    description="Generate captions and questions for images using pre-trained models.",
+    theme=seafoam,
 )
 
-gr_interface.launch()
+# Launch the interface
+gr_interface.launch(share=True)