Update download_dataset.py

download_dataset.py

@@ -1,17 +1,17 @@
 """
-download_dataset.py — Tải và xử lý ViLAP dataset từ YouTube.
+download_dataset.py — Download and process the ViLAP dataset from YouTube.
 
-Yêu cầu:
+Requirements:
     pip install yt-dlp mediapipe opencv-python librosa scipy numpy
 
-Cách dùng:
-    # Tải tất cả (train + val + test)
+Usage:
+    # Download all splits (train + val + test)
     python download_dataset.py --output ./vilap_data
 
-    # Chỉ tải test split
+    # Download only the test split
     python download_dataset.py --output ./vilap_data --splits test
 
-    # Chỉ tải một số videos cụ thể
+    # Download only specific videos
     python download_dataset.py --output ./vilap_data --video_ids 3p7dFrIx5bk xwsPD6xiPbI
 """
 
@@ -33,7 +33,7 @@ def load_sources():
 
 
 def clips_for_splits(splits):
-    """Trả về dict {video_id: [clip_ids]} cho các split được chọn."""
+    """Return a dictionary {video_id: [clip_ids]} for the selected splits."""
     clips = {}
     for split in splits:
         path = FILELISTS_DIR / f'{split}.txt'
@@ -50,7 +50,7 @@ def clips_for_splits(splits):
 
 # ─── Step 1: Download raw video ───────────────────────────────────────────────
 def download_video(url, out_path):
-    """Tải video với yt-dlp, chất lượng tốt nhất dưới 1080p."""
+    """Download a video with yt-dlp using the best available quality below 1080p."""
     cmd = [
         'yt-dlp', '-f', 'bestvideo[height<=720][ext=mp4]+bestaudio[ext=m4a]/best[height<=720][ext=mp4]/best',
         '--merge-output-format', 'mp4',
@@ -64,9 +64,9 @@ def download_video(url, out_path):
 
 # ─── Step 2: Cut clips ────────────────────────────────────────────────────────
 def cut_clips(raw_video, clips_dir, clip_ids, fps=25):
-    """Cắt video thành clips 5 giây theo clip index."""
+    """Cut the video into 5-second clips based on the clip index."""
     clips_dir.mkdir(parents=True, exist_ok=True)
-    # Lấy duration
+    # Get the video duration.
     probe = subprocess.run(
         ['ffprobe', '-v', 'error', '-show_entries', 'format=duration',
          '-of', 'default=noprint_wrappers=1:nokey=1', str(raw_video)],
@@ -78,7 +78,7 @@ def cut_clips(raw_video, clips_dir, clip_ids, fps=25):
         print(f'  [WARN] cannot get duration for {raw_video.name}')
         return
 
-    # Đọc clip indices từ tên như "spk_c0017" → index 17
+    # Read the clip index from names such as "spk_c0017" → index 17.
     for clip_id in clip_ids:
         idx = int(clip_id.split('_c')[-1])
         start_sec = idx * 5.0
@@ -99,7 +99,7 @@ def cut_clips(raw_video, clips_dir, clip_ids, fps=25):
 
 # ─── Step 3: Extract faces ────────────────────────────────────────────────────
 def extract_faces(clips_dir, face_dir, clip_id):
-    """Dùng MediaPipe Face Detection để crop và resize khuôn mặt 128x128."""
+    """Use MediaPipe Face Detection to crop and resize the face to 128x128."""
     try:
         import cv2, mediapipe as mp
     except ImportError:
@@ -129,7 +129,7 @@ def extract_faces(clips_dir, face_dir, clip_id):
             y1 = max(0, int(bb.ymin * h))
             x2 = min(w, int((bb.xmin + bb.width) * w))
             y2 = min(h, int((bb.ymin + bb.height) * h))
-            # Thêm margin 20%
+            # Add a 20% margin around the detected face.
             margin_x = int((x2 - x1) * 0.2)
             margin_y = int((y2 - y1) * 0.2)
             x1 = max(0, x1 - margin_x); y1 = max(0, y1 - margin_y)
@@ -145,7 +145,7 @@ def extract_faces(clips_dir, face_dir, clip_id):
 
 # ─── Step 4: Extract landmarks ───────────────────────────────────────────────
 def extract_landmarks(face_dir, lm_dir, clip_id):
-    """Dùng MediaPipe Face Mesh để extract 478 landmarks, lưu pose + content."""
+    """Use MediaPipe Face Mesh to extract 478 landmarks and save pose + content landmarks."""
     try:
         import cv2, mediapipe as mp
         import numpy as np
@@ -166,11 +166,11 @@ def extract_landmarks(face_dir, lm_dir, clip_id):
     CONTENT_IDS = [
         61,185,40,39,37,0,267,269,270,409,291,375,321,405,314,17,84,181,91,146,
         78,191,80,81,82,13,312,311,310,415,308,324,318,402,317,14,87,178,88,95,
-        # jaw (inner portion)
+        # Jaw landmarks from the inner portion.
         162,127,234,93,132,58,172,136,150,149,176,148,152,
         377,400,378,379,
     ]
-    # Dùng đúng 57 content points theo IP-LAP convention
+    # Use exactly 57 content points following the IP-LAP convention.
     CONTENT_IDS = [
         61,185,40,39,37,0,267,269,270,409,291,375,321,405,314,17,84,181,91,146,
         78,191,80,81,82,13,312,311,310,415,308,324,318,402,317,14,87,178,88,95,
@@ -207,7 +207,7 @@ def extract_landmarks(face_dir, lm_dir, clip_id):
 
 # ─── Step 5: Extract audio ───────────────────────────────────────────────────
 def extract_audio(clips_dir, audio_dir, clip_id):
-    """Trích xuất WAV + mel spectrogram theo IP-LAP convention."""
+    """Extract WAV audio and mel spectrograms following the IP-LAP convention."""
     try:
         import numpy as np, librosa
         from scipy import signal as scipy_signal
@@ -228,7 +228,7 @@ def extract_audio(clips_dir, audio_dir, clip_id):
     if not wav_path.exists():
         return False
 
-    # Mel spectrogram theo IP-LAP convention
+    # Compute the mel spectrogram following the IP-LAP convention.
     wav, _ = librosa.load(str(wav_path), sr=AUDIO_SR)
     preemph = scipy_signal.lfilter([1, -0.97], [1], wav)
     D = librosa.stft(preemph, n_fft=800, hop_length=200, win_length=800)
@@ -250,7 +250,7 @@ def process_video(video_id, url, clip_ids, out_base):
 
     raw_video = raw_dir / f'{video_id}.mp4'
 
-    # Step 1: Download
+    # Step 1: Download the raw video.
     if not raw_video.exists():
         print(f'  Downloading {url}...')
         if not download_video(url, raw_video):
@@ -259,7 +259,7 @@ def process_video(video_id, url, clip_ids, out_base):
     else:
         print(f'  [SKIP] already downloaded')
 
-    # Step 2–5: Per-clip
+    # Steps 2–5: Process each clip.
     for clip_id in clip_ids:
         full_clip_id = clip_id   # e.g. "3p7dFrIx5bk_c0017"
         clip_path = clips_dir / f'{full_clip_id}.mp4'
@@ -299,7 +299,7 @@ def main():
     sources = load_sources()
     clips_map = clips_for_splits(args.splits)
 
-    # Copy filelists
+    # Copy filelists to the output directory.
     fl_out = out_base / 'filelists'
     fl_out.mkdir(exist_ok=True)
     for split in args.splits:
@@ -307,7 +307,7 @@ def main():
         if src.exists():
             shutil.copy(src, fl_out / f'{split}.txt')
 
-    # Filter by video_ids if specified
+    # Filter by video_ids if specified.
     if args.video_ids:
         clips_map = {k: v for k, v in clips_map.items() if k in args.video_ids}
 
@@ -338,4 +338,4 @@ def main():
 
 
 if __name__ == '__main__':
-    main()
+    main()