Update app.py

app.py

@@ -1,5 +1,5 @@
 import streamlit as st
-from transformers import AutoProcessor, Qwen2VLForConditionalGeneration
+from transformers import AutoProcessor, Qwen2VLForConditionalGeneration, AutoModelForCausalLM, AutoTokenizer
 from PIL import Image
 import torch
 import cv2
@@ -9,89 +9,123 @@ import tempfile
 processor = AutoProcessor.from_pretrained("Qwen/Qwen2-VL-2B-Instruct")
 model = Qwen2VLForConditionalGeneration.from_pretrained("Qwen/Qwen2-VL-2B-Instruct")
 
+# Load Meta-Llama model and tokenizer for story generation
+llama_tokenizer = AutoTokenizer.from_pretrained("meta-llama/Meta-Llama-3.1-8B-Instruct")
+llama_model = AutoModelForCausalLM.from_pretrained("meta-llama/Meta-Llama-3.1-8B-Instruct")
+
 # Check if CUDA is available and set the device accordingly
-device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+device = torch.device("cpu")
 model.to(device)
+llama_model.to(device)
 
 # Streamlit app
 st.title("Media Description Generator")
 
-uploaded_file = st.file_uploader("Choose an image or video...", type=["jpg", "jpeg", "png", "mp4", "avi", "mov"])
-
-if uploaded_file is not None:
-    file_type = uploaded_file.type.split('/')[0]
-
-    if file_type == 'image':
-        # Open the image
-        image = Image.open(uploaded_file)
-        st.image(image, caption='Uploaded Image.', use_column_width=True)
-        st.write("Generating description...")
-
-    elif file_type == 'video':
-        # Save the uploaded video to a temporary file
-        tfile = tempfile.NamedTemporaryFile(delete=False)
-        tfile.write(uploaded_file.read())
-
-        # Open the video file
-        cap = cv2.VideoCapture(tfile.name)
-
-        # Extract the first frame
-        ret, frame = cap.read()
-        if not ret:
-            st.error("Failed to read the video file.")
-            st.stop()
-        else:
-            # Convert the frame to an image
-            image = Image.fromarray(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))
-            st.image(image, caption='First Frame of Uploaded Video.', use_column_width=True)
-            st.write("Generating description...")
-
-        # Release the video capture object
-        cap.release()
-
-    else:
-        st.error("Unsupported file type.")
-        st.stop()
+uploaded_files = st.file_uploader("Choose images or videos...", type=["jpg", "jpeg", "png", "mp4", "avi", "mov"], accept_multiple_files=True)
 
-    # Add a text input for the user to ask a question
-    user_question = st.text_input("Ask a question about the image or video:")
+if uploaded_files:
+    user_question = st.text_input("Ask a question about the images or videos:")
 
     if user_question:
-        messages = [
-            {
-                "role": "user",
-                "content": [
-                    {
-                        "type": "image",
-                        "image": image,
-                    },
-                    {"type": "text", "text": user_question},
-                ],
-            }
-        ]
-
-        # Preparation for inference
-        text = processor.apply_chat_template(
-            messages, tokenize=False, add_generation_prompt=True
-        )
-
-        # Pass the image to the processor
-        inputs = processor(
-            text=[text],
-            images=[image],
-            padding=True,
-            return_tensors="pt",
-        )
-        inputs = inputs.to(device)  # Ensure inputs are on the same device as the model
-
-        # Inference: Generation of the output
-        generated_ids = model.generate(**inputs, max_new_tokens=128)
-        generated_ids_trimmed = [
-            out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
-        ]
-        output_text = processor.batch_decode(
-            generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
-        )
-        
-        st.write("Description:")
-        st.write(output_text[0])
+        all_output_texts = []  # Initialize an empty list to store all output texts
+
+        for uploaded_file in uploaded_files:
+            file_type = uploaded_file.type.split('/')[0]
+
+            if file_type == 'image':
+                # Open the image
+                image = Image.open(uploaded_file)
+                # Resize image to reduce memory usage
+                image = image.resize((512, 512))
+                st.image(image, caption='Uploaded Image.', use_column_width=True)
+                st.write("Generating description...")
+
+            elif file_type == 'video':
+                # Save the uploaded video to a temporary file
+                tfile = tempfile.NamedTemporaryFile(delete=False)
+                tfile.write(uploaded_file.read())
+
+                # Open the video file
+                cap = cv2.VideoCapture(tfile.name)
+
+                # Extract the first frame
+                ret, frame = cap.read()
+                if not ret:
+                    st.error("Failed to read the video file.")
+                    continue
+                else:
+                    # Convert the frame to an image
+                    image = Image.fromarray(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))
+                    # Resize image to reduce memory usage
+                    image = image.resize((512, 512))
+                    st.image(image, caption='First Frame of Uploaded Video.', use_column_width=True)
+                    st.write("Generating description...")
+
+                # Release the video capture object
+                cap.release()
+
+            else:
+                st.error("Unsupported file type.")
+                continue
+
+            messages = [
+                {
+                    "role": "user",
+                    "content": [
+                        {
+                            "type": "image",
+                            "image": image,
+                        },
+                        {"type": "text", "text": user_question},
+                    ],
+                }
+            ]
+
+            # Preparation for inference
+            text = processor.apply_chat_template(
+                messages, tokenize=False, add_generation_prompt=True
+            )
+
+            # Pass the image to the processor
+            inputs = processor(
+                text=[text],
+                images=[image],
+                padding=True,
+                return_tensors="pt",
+            )
+            inputs = inputs.to(device)  # Ensure inputs are on the same device as the model
+
+            # Inference: Generation of the output
+            generated_ids = model.generate(**inputs, max_new_tokens=128)
+            generated_ids_trimmed = [
+                out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
+            ]
+            output_text = processor.batch_decode(
+                generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
+            )
+
+            st.write("Description:")
+            st.write(output_text[0])
+
+            # Append the output text to the list
+            all_output_texts.append(output_text[0])
+
+            # Clear memory after processing each file
+            del image, inputs, generated_ids, generated_ids_trimmed, output_text
+            torch.cuda.empty_cache()
+            torch.manual_seed(0)  # Reset the seed to ensure reproducibility
+
+        # Combine all descriptions into a single text
+        combined_text = " ".join(all_output_texts)
+
+        # Create a custom prompt
+        custom_prompt = f"Based on the following descriptions, create a short story:\n\n{combined_text}\n\nStory:"
+
+        # Generate a story using Meta-Llama
+        inputs = llama_tokenizer.encode(custom_prompt, return_tensors="pt").to(device)
+        story_ids = llama_model.generate(inputs, max_length=500, num_return_sequences=1)
+        story = llama_tokenizer.decode(story_ids[0], skip_special_tokens=True)
+
+        # Display the generated story
+        st.write("Generated Story:")
+        st.write(story)