Update app.py

app.py

@@ -1,112 +1,79 @@
-import sys
-import subprocess
-import pkg_resources
-import os
-import tempfile
-import requests
-
-required_packages = {
-    'torch': 'torch',
-    'gradio': 'gradio',
-    'transformers': 'transformers',
-    'decord': 'decord',
-    'numpy': 'numpy',
-    'instaloader': 'instaloader'
-}
-
-def install_packages(packages):
-    for package in packages:
-        subprocess.check_call([sys.executable, "-m", "pip", "install", package])
-
-def check_and_install_packages():
-    installed_packages = {pkg.key for pkg in pkg_resources.working_set}
-    missing_packages = [required_packages[pkg] for pkg in required_packages if pkg not in installed_packages]
-    
-    if missing_packages:
-        print("Installing missing packages...")
-        install_packages(missing_packages)
-        print("Packages installed successfully.")
-    else:
-        print("All required packages are already installed.")
-
-# Check and install required packages
-check_and_install_packages()
-
-# Now import the required modules
 import gradio as gr
 import torch
-from transformers import AutoTokenizer, AutoModelForCausalLM
+from llava.model.builder import load_pretrained_model
+from llava.mm_utils import get_model_name_from_path, process_images, tokenizer_image_token
+from llava.constants import IMAGE_TOKEN_INDEX, DEFAULT_IMAGE_TOKEN
+from llava.conversation import conv_templates
+import copy
 from decord import VideoReader, cpu
 import numpy as np
-import instaloader
-
-# Initialize Instaloader
-L = instaloader.Instaloader()
-
-def download_instagram_video(url):
-    try:
-        # Use a third-party API service like savefrom.net
-        api_url = f"https://api.savefrom.net/1/info?url={url}"
-        response = requests.get(api_url)
-        data = response.json()
-
-        if 'url' in data:
-            video_url = data['url']
-            video_data = requests.get(video_url)
 
-            # Save the video file locally
-            with open('downloaded_reel.mp4', 'wb') as f:
-                f.write(video_data.content)
+title = "# 📸 Instagram Reels Analiz Aracı"
+description = """Bu araç, yüklenen Instagram Reels videolarını analiz eder ve içeriği özetler. 
+Video hakkında genel bir açıklama yapar ve klipte neler olup bittiğini adım adım anlatır."""
 
-            return 'downloaded_reel.mp4'
-        else:
-            print(f"Error: {data.get('error', 'Unknown error')}")
-            return None
-    except Exception as e:
-        print(f"Error downloading video: {e}")
-        return None
-
-def process_video(video_path, max_frames=64):
+def load_video(video_path, max_frames_num=64, fps=1):
     vr = VideoReader(video_path, ctx=cpu(0))
-    total_frames = len(vr)
-    frame_indices = np.linspace(0, total_frames - 1, max_frames, dtype=int)
-    frames = vr.get_batch(frame_indices).asnumpy()
-    return frames
+    total_frame_num = len(vr)
+    frame_idx = list(range(0, total_frame_num, int(vr.get_avg_fps() / fps)))
+    
+    if len(frame_idx) > max_frames_num:
+        frame_idx = np.linspace(0, total_frame_num - 1, max_frames_num, dtype=int).tolist()
+    
+    video_frames = vr.get_batch(frame_idx).asnumpy()
+    return video_frames, len(frame_idx)
+
+# Model yükleme
+pretrained = "lmms-lab/LLaVA-Video-7B-Qwen2"
+model_name = "llava_qwen"
+device = "cuda" if torch.cuda.is_available() else "cpu"
 
-# This is a placeholder for the actual LLaVA-Video model
-def analyze_video(video_frames, question):
-    # In a real implementation, you would use the LLaVA-Video model here
-    return f"Analyzed {len(video_frames)} frames. Your question was: {question}"
+print("Model yükleniyor...")
+tokenizer, model, image_processor, max_length = load_pretrained_model(pretrained, None, model_name, torch_dtype="bfloat16", device_map="auto")
+model.eval()
+print("Model başarıyla yüklendi!")
 
-def analyze_instagram_video(video_url, question):
-    if not video_url:
-        return "Please enter an Instagram video URL."
+def analyze_reel(video_path):
+    video_frames, frame_count = load_video(video_path)
+    video = image_processor.preprocess(video_frames, return_tensors="pt")["pixel_values"].to(device).bfloat16()
     
-    video_path = download_instagram_video(video_url)
-    if not video_path:
-        return "Failed to download the video. Please check the URL and try again."
+    prompt = f"{DEFAULT_IMAGE_TOKEN}Bu Instagram Reels videosunu analiz et. Önce videonun genel içeriğini özetle, ardından klipte neler olup bittiğini adım adım açıkla. Video {frame_count} kareye bölünmüştür."
     
-    video_frames = process_video(video_path)
-    response = analyze_video(video_frames, question)
+    conv = copy.deepcopy(conv_templates["qwen_1_5"])
+    conv.append_message(conv.roles[0], prompt)
+    conv.append_message(conv.roles[1], None)
+    prompt = conv.get_prompt()
+    
+    input_ids = tokenizer_image_token(prompt, tokenizer, IMAGE_TOKEN_INDEX, return_tensors="pt").unsqueeze(0).to(device)
+    
+    with torch.no_grad():
+        output = model.generate(
+            input_ids,
+            images=[video],
+            modalities=["video"],
+            do_sample=False,
+            temperature=0,
+            max_new_tokens=1024,
+        )
+    
+    response = tokenizer.batch_decode(output, skip_special_tokens=True)[0].strip()
     return response
 
-# Create Gradio interface
+def gradio_interface(video_file):
+    if video_file is None:
+        return "Lütfen bir video dosyası yükleyin."
+    return analyze_reel(video_file)
+
 with gr.Blocks() as demo:
-    gr.Markdown("# 🎥 Instagram Video Analyzer")
-    gr.Markdown("Enter the URL of an Instagram video and ask questions about its content!")
+    gr.Markdown(title)
+    gr.Markdown(description)
     
     with gr.Row():
-        with gr.Column():
-            video_url_input = gr.Textbox(label="Instagram Video URL", placeholder="https://www.instagram.com/p/...")
-            question_input = gr.Textbox(label="Ask a question about the video", placeholder="What's happening in this Instagram video?")
-            submit_button = gr.Button("Analyze Video")
-        output = gr.Textbox(label="Analysis Result")
+        video_input = gr.Video(label="Instagram Reels Videosu")
+        output = gr.Textbox(label="Analiz Sonucu", lines=10)
     
-    submit_button.click(
-        fn=analyze_instagram_video,
-        inputs=[video_url_input, question_input],
-        outputs=output
-    )
+    analyze_button = gr.Button("Reels'i Analiz Et")
+    analyze_button.click(fn=gradio_interface, inputs=video_input, outputs=output)
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch(share=True)