syncnet-python 0.2.0__tar.gz → 0.2.1__tar.gz

{syncnet_python-0.2.0 → syncnet_python-0.2.1}/INSTALL.md

@@ -2,7 +2,7 @@
 
 ## Prerequisites
 
-- Python 3.13 or higher
+- Python 3.9 or higher
 - CUDA-capable GPU (optional, but recommended)
 - FFmpeg
 

{syncnet_python-0.2.0 → syncnet_python-0.2.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: syncnet-python
-Version: 0.2.0
+Version: 0.2.1
 Summary: SyncNet: Audio-visual synchronization detection using deep learning. Updated version of https://github.com/joonson/syncnet_python for modern Python versions.
 Author: SyncNet Python Contributors
 Maintainer: SyncNet Python Contributors
@@ -47,6 +47,11 @@ Dynamic: license-file
 
 # SyncNet Python
 
+[![PyPI version](https://badge.fury.io/py/syncnet-python.svg)](https://badge.fury.io/py/syncnet-python)
+[![Python](https://img.shields.io/pypi/pyversions/syncnet-python.svg)](https://pypi.org/project/syncnet-python/)
+[![Downloads](https://pepy.tech/badge/syncnet-python)](https://pepy.tech/project/syncnet-python)
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
+
 Audio-visual synchronization detection using deep learning with modern Python architecture.
 
 This is a **refactored and enhanced version** of the original [SyncNet implementation](https://github.com/joonson/syncnet_python) by Joon Son Chung, updated for Python 3.9+ with clean architecture, comprehensive error handling, and performance optimizations.

{syncnet_python-0.2.0 → syncnet_python-0.2.1}/README.md

@@ -1,5 +1,10 @@
 # SyncNet Python
 
+[![PyPI version](https://badge.fury.io/py/syncnet-python.svg)](https://badge.fury.io/py/syncnet-python)
+[![Python](https://img.shields.io/pypi/pyversions/syncnet-python.svg)](https://pypi.org/project/syncnet-python/)
+[![Downloads](https://pepy.tech/badge/syncnet-python)](https://pepy.tech/project/syncnet-python)
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
+
 Audio-visual synchronization detection using deep learning with modern Python architecture.
 
 This is a **refactored and enhanced version** of the original [SyncNet implementation](https://github.com/joonson/syncnet_python) by Joon Son Chung, updated for Python 3.9+ with clean architecture, comprehensive error handling, and performance optimizations.

{syncnet_python-0.2.0 → syncnet_python-0.2.1}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "syncnet-python"
-version = "0.2.0"
+version = "0.2.1"
 description = "SyncNet: Audio-visual synchronization detection using deep learning. Updated version of https://github.com/joonson/syncnet_python for modern Python versions."
 readme = "README.md"
 requires-python = ">=3.9"

{syncnet_python-0.2.0 → syncnet_python-0.2.1}/syncnet_python/syncnet_pipeline.py

@@ -18,10 +18,17 @@ from scipy.interpolate import interp1d
 from scenedetect import ContentDetector, SceneManager, StatsManager
 from scenedetect.video_manager import VideoManager
 
-from .detectors.s3fd import S3FD
-from .detectors.s3fd.nets import S3FDNet
-from .SyncNetInstance import SyncNetInstance
-from .SyncNetModel import S
+try:
+    from .detectors.s3fd import S3FD
+    from .detectors.s3fd.nets import S3FDNet
+    from .SyncNetInstance import SyncNetInstance
+    from .SyncNetModel import S
+except ImportError:
+    # Fallback for direct script execution
+    from detectors.s3fd import S3FD
+    from detectors.s3fd.nets import S3FDNet
+    from SyncNetInstance import SyncNetInstance
+    from SyncNetModel import S
 
 # ---------------------------------------------------------------------- #
 # Configuration                                                          #
@@ -47,6 +54,11 @@ class PipelineConfig:
     # Tools
     ffmpeg_bin: str = "ffmpeg"  # assumes ffmpeg in $PATH
     audio_sample_rate: int = 16000  # resample rate for speech
+    
+    def __post_init__(self):
+        """Validate configuration after initialization."""
+        if self.ffmpeg_bin is None:
+            self.ffmpeg_bin = "ffmpeg"
 
     @classmethod
     def from_dict(cls, d: Dict[str, Any]):
@@ -165,16 +177,41 @@ class SyncNetPipeline:
         slice_wav = f"{base}.wav"
         ss = track["frame"][0] / cfg.frame_rate
         to = (track["frame"][-1] + 1) / cfg.frame_rate
-        subprocess.call(
-            f'{cfg.ffmpeg_bin} -y -i "{audio_wav}" -ss {ss:.3f} -to {to:.3f} "{slice_wav}"',
-            shell=True,
-        )
+        
+        # Ensure ffmpeg_bin is not None
+        ffmpeg_bin = cfg.ffmpeg_bin if cfg.ffmpeg_bin is not None else "ffmpeg"
+        
+        cmd = [
+            ffmpeg_bin, "-y", "-i", str(audio_wav), 
+            "-ss", f"{ss:.3f}", "-to", f"{to:.3f}", 
+            str(slice_wav)
+        ]
+        
+        try:
+            result = subprocess.run(cmd, capture_output=True, text=True, check=True)
+        except subprocess.CalledProcessError as e:
+            logging.error(f"FFmpeg audio slicing failed: {e.stderr}")
+            raise RuntimeError(f"FFmpeg audio slicing failed: {e.stderr}")
+        except FileNotFoundError:
+            logging.error(f"FFmpeg not found at: {ffmpeg_bin}")
+            raise RuntimeError(f"FFmpeg not found. Please ensure ffmpeg is installed and in PATH.")
 
         final_avi = f"{base}.avi"
-        subprocess.call(
-            f'{cfg.ffmpeg_bin} -y -i "{tmp_avi}" -i "{slice_wav}" -c:v copy -c:a copy "{final_avi}"',
-            shell=True,
-        )
+        
+        cmd = [
+            ffmpeg_bin, "-y", "-i", str(tmp_avi), "-i", str(slice_wav),
+            "-c:v", "copy", "-c:a", "copy", str(final_avi)
+        ]
+        
+        try:
+            result = subprocess.run(cmd, capture_output=True, text=True, check=True)
+        except subprocess.CalledProcessError as e:
+            logging.error(f"FFmpeg video/audio merge failed: {e.stderr}")
+            raise RuntimeError(f"FFmpeg video/audio merge failed: {e.stderr}")
+        except FileNotFoundError:
+            logging.error(f"FFmpeg not found at: {ffmpeg_bin}")
+            raise RuntimeError(f"FFmpeg not found. Please ensure ffmpeg is installed and in PATH.")
+        
         os.remove(tmp_avi)
         return final_avi
 

syncnet_python-0.2.1/syncnet_python/test_lse_metrics.py

@@ -0,0 +1,95 @@
+import logging
+import sys
+from pathlib import Path
+from syncnet_pipeline import SyncNetPipeline
+
+logging.basicConfig(
+    level=logging.INFO,
+    format="%(asctime)s [%(levelname)s] %(message)s"
+)
+
+def calculate_lse_metrics(video_path):
+    """
+    Calculate LSE-C and LSE-D metrics for a video
+    
+    LSE-C (Lip Sync Error - Confidence): 
+    - Higher confidence values indicate better lip sync
+    - Threshold typically around 3.5-4.0 for good sync
+    
+    LSE-D (Lip Sync Error - Distance):
+    - Lower distance values indicate better lip sync  
+    - Threshold typically around 6.5-7.0 for good sync
+    """
+    # Initialize pipeline
+    pipe = SyncNetPipeline(
+        {
+            "s3fd_weights": "../weights/sfd_face.pth",
+            "syncnet_weights": "../weights/syncnet_v2.model",
+        },
+        device="cuda",  # or "cpu"
+    )
+    
+    # Run inference
+    print(f"\n=== Processing {video_path} ===")
+    
+    # For testing, we'll extract audio from the video itself
+    results = pipe.inference(
+        video_path=video_path,
+        audio_path=video_path,  # Extract audio from same video
+        cache_dir="../example/cache_lse",
+    )
+    
+    offsets, confs, dists, max_conf, min_dist, s3fd_json, has_face = results
+    
+    if not has_face:
+        print(f"No face detected in {video_path}")
+        return None, None
+    
+    # LSE-C is the maximum confidence across all tracks
+    lse_c = max_conf
+    
+    # LSE-D is the minimum distance across all tracks
+    lse_d = min_dist
+    
+    print(f"Number of face tracks: {len(offsets)}")
+    print(f"Track offsets: {offsets}")
+    print(f"Track confidences: {[f'{c:.3f}' for c in confs]}")
+    print(f"Track distances: {[f'{d:.3f}' for d in dists]}")
+    print(f"LSE-C (max confidence): {lse_c:.3f}")
+    print(f"LSE-D (min distance): {lse_d:.3f}")
+    
+    # Interpretation
+    sync_quality = "GOOD" if lse_c > 3.5 and lse_d < 7.0 else "POOR"
+    print(f"Sync Quality: {sync_quality}")
+    
+    return lse_c, lse_d
+
+def main():
+    # Test files
+    test_files = [
+        "../example/pair_0000_lipsynced.mp4",
+        "../example/pair_0001_lipsynced.mp4"
+    ]
+    
+    results = {}
+    
+    for video_path in test_files:
+        if Path(video_path).exists():
+            lse_c, lse_d = calculate_lse_metrics(video_path)
+            results[video_path] = (lse_c, lse_d)
+        else:
+            print(f"File not found: {video_path}")
+    
+    # Summary
+    print("\n=== SUMMARY ===")
+    print(f"{'Video':<40} {'LSE-C':<10} {'LSE-D':<10} {'Quality':<10}")
+    print("-" * 70)
+    
+    for video_path, (lse_c, lse_d) in results.items():
+        if lse_c is not None:
+            video_name = Path(video_path).name
+            quality = "GOOD" if lse_c > 3.5 and lse_d < 7.0 else "POOR"
+            print(f"{video_name:<40} {lse_c:<10.3f} {lse_d:<10.3f} {quality:<10}")
+
+if __name__ == "__main__":
+    main()

{syncnet_python-0.2.0 → syncnet_python-0.2.1}/syncnet_python.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: syncnet-python
-Version: 0.2.0
+Version: 0.2.1
 Summary: SyncNet: Audio-visual synchronization detection using deep learning. Updated version of https://github.com/joonson/syncnet_python for modern Python versions.
 Author: SyncNet Python Contributors
 Maintainer: SyncNet Python Contributors
@@ -47,6 +47,11 @@ Dynamic: license-file
 
 # SyncNet Python
 
+[![PyPI version](https://badge.fury.io/py/syncnet-python.svg)](https://badge.fury.io/py/syncnet-python)
+[![Python](https://img.shields.io/pypi/pyversions/syncnet-python.svg)](https://pypi.org/project/syncnet-python/)
+[![Downloads](https://pepy.tech/badge/syncnet-python)](https://pepy.tech/project/syncnet-python)
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
+
 Audio-visual synchronization detection using deep learning with modern Python architecture.
 
 This is a **refactored and enhanced version** of the original [SyncNet implementation](https://github.com/joonson/syncnet_python) by Joon Son Chung, updated for Python 3.9+ with clean architecture, comprehensive error handling, and performance optimizations.

{syncnet_python-0.2.0 → syncnet_python-0.2.1}/syncnet_python.egg-info/SOURCES.txt

@@ -44,6 +44,7 @@ syncnet_python/run_syncnet_pipeline_on_1example.py
 syncnet_python/run_syncnet_pipeline_on_mocha_generation_on_mocha_bench.py
 syncnet_python/run_syncnet_pipeline_on_your_own_model_results.py
 syncnet_python/syncnet_pipeline.py
+syncnet_python/test_lse_metrics.py
 syncnet_python.egg-info/PKG-INFO
 syncnet_python.egg-info/SOURCES.txt
 syncnet_python.egg-info/dependency_links.txt