remove cuda ffmpeg

Dockerfile

@@ -23,8 +23,6 @@ RUN apt-get update && \
     liblzma-dev \
     # gradio dependencies \
     ffmpeg \
-    # NVIDIA Video Codec SDK \
-    libnvidia-encode-12-3 \
     && apt-get clean \
     && rm -rf /var/lib/apt/lists/*
 

src/video_processor/processor.py

@@ -1,6 +1,6 @@
 import torch
 from transformers import AutoProcessor, AutoModelForImageTextToText
-from typing import List, Dict, Generator
+from typing import List, Dict
 import logging
 import os
 import subprocess
@@ -103,7 +103,7 @@ Be specific about visual details but stay concise."""}
         )
         return self.processor.batch_decode(outputs, skip_special_tokens=True)[0].split("Assistant: ")[-1]
 
-    def process_video(self, video_path: str, segment_length: int = 10) -> Generator[Dict, None, None]:
+    def process_video(self, video_path: str, segment_length: int = 10) -> List[Dict]:
         try:
             # Create temp directory for segments
             temp_dir = tempfile.mkdtemp()
@@ -128,49 +128,22 @@ Be specific about visual details but stay concise."""}
                 cmd = [
                     "ffmpeg",
                     "-y",  # Overwrite output files
-                    "-hwaccel", "cuda",  # Use CUDA hardware acceleration
-                    "-hwaccel_output_format", "cuda",  # Keep frames in GPU memory
-                    "-threads", "0",  # Use all available CPU threads
-                    "-thread_type", "frame",  # Frame-level multi-threading
+                    "-threads", "4",  # Use 4 threads
                     "-i", video_path,
                     "-ss", str(start_time),  # Seek position
                     "-t", str(end_time - start_time),  # Duration
-                    "-c:v", "h264_nvenc",  # Use NVIDIA hardware encoder
-                    "-preset", "p1",  # Lowest latency preset for NVENC
-                    "-tune", "ll",  # Low latency tuning
-                    "-rc", "vbr",  # Variable bitrate mode
-                    "-cq", "28",  # Quality-based VBR
-                    "-b:v", "0",  # Let VBR control bitrate
-                    "-vf", "scale_cuda=640:-2",  # GPU-accelerated scaling
+                    "-c:v", "libx264",  # Video codec
+                    "-preset", "ultrafast",
+                    "-tune", "fastdecode",
+                    "-crf", "28",  # Lower quality but faster
+                    "-vf", "scale=640:-2",  # Resize to smaller resolution
                     "-an",  # Remove audio
+                    "-pix_fmt", "yuv420p",
                     segment_path
                 ]
                 
                 ffmpeg_start = time.time()
-                try:
-                    result = subprocess.run(cmd, check=True, capture_output=True, text=True)
-                    logger.debug(f"FFmpeg output: {result.stderr}")
-                except subprocess.CalledProcessError as e:
-                    logger.error(f"FFmpeg error: {e.stderr}")
-                    # Fallback to CPU if GPU encoding fails
-                    logger.warning("Falling back to CPU encoding")
-                    cmd = [
-                        "ffmpeg",
-                        "-y",
-                        "-threads", "0",
-                        "-i", video_path,
-                        "-ss", str(start_time),
-                        "-t", str(end_time - start_time),
-                        "-c:v", "libx264",
-                        "-preset", "ultrafast",
-                        "-tune", "fastdecode",
-                        "-crf", "28",
-                        "-vf", "scale=640:-2",
-                        "-an",
-                        "-pix_fmt", "yuv420p",
-                        segment_path
-                    ]
-                    subprocess.run(cmd, check=True, capture_output=True)
+                subprocess.run(cmd, check=True, capture_output=True)
                 ffmpeg_time = time.time() - ffmpeg_start
                 
                 # Analyze segment