Update core/clip_extractor.py

core/clip_extractor.py

@@ -123,14 +123,16 @@ class ClipExtractor:
         candidates: List[ClipCandidate],
         num_clips: int,
         enforce_diversity: bool = True,
+        video_duration: Optional[float] = None,
     ) -> List[ClipCandidate]:
         """
-        Select top clips from candidates.
+        Select top clips from candidates with adaptive diversity.
 
         Args:
             candidates: List of clip candidates with scores
             num_clips: Number of clips to select
             enforce_diversity: Enforce minimum gap between clips
+            video_duration: Total video duration for adaptive gap calculation
 
         Returns:
             List of selected ClipCandidate objects
@@ -147,9 +149,23 @@ class ClipExtractor:
         if not enforce_diversity:
             return sorted_candidates[:num_clips]
 
+        # Calculate adaptive minimum gap based on video duration
+        # For longer videos, we want clips spread across the entire video
+        if video_duration and video_duration > 0:
+            # Target: clips should be spread across the video
+            # Divide video into (num_clips + 1) sections, min_gap is section size
+            adaptive_gap = video_duration / (num_clips + 1)
+            # But don't go below config minimum or above 1/3 of video
+            min_gap = max(
+                self.config.min_gap_between_clips,
+                min(adaptive_gap, video_duration / 3)
+            )
+            logger.debug(f"Adaptive min_gap: {min_gap:.1f}s for {video_duration:.1f}s video")
+        else:
+            min_gap = self.config.min_gap_between_clips
+
         # Select with diversity constraint
         selected = []
-        min_gap = self.config.min_gap_between_clips
 
         for candidate in sorted_candidates:
             if len(selected) >= num_clips:
@@ -167,11 +183,33 @@ class ClipExtractor:
             if is_diverse:
                 selected.append(candidate)
 
-        # If we couldn't get enough with diversity, relax constraint
+        # If we couldn't get enough with strict diversity, progressively relax
+        if len(selected) < num_clips:
+            # Try with 50% of the gap
+            relaxed_gap = min_gap * 0.5
+            logger.info(f"Relaxing diversity to {relaxed_gap:.1f}s gap")
+
+            for candidate in sorted_candidates:
+                if len(selected) >= num_clips:
+                    break
+                if candidate in selected:
+                    continue
+
+                is_diverse = True
+                for existing in selected:
+                    gap = abs(candidate.start_time - existing.start_time)
+                    if gap < relaxed_gap:
+                        is_diverse = False
+                        break
+
+                if is_diverse:
+                    selected.append(candidate)
+
+        # Final fallback: just take top clips if still not enough
         if len(selected) < num_clips:
             logger.warning(
                 f"Only {len(selected)} diverse clips found, "
-                f"relaxing diversity constraint"
+                f"filling remaining from top candidates"
             )
             for candidate in sorted_candidates:
                 if candidate not in selected:
@@ -179,7 +217,7 @@ class ClipExtractor:
                 if len(selected) >= num_clips:
                     break
 
-        logger.info(f"Selected {len(selected)} clips from {len(candidates)} candidates")
+        logger.info(f"Selected {len(selected)} clips from {len(candidates)} candidates (gap={min_gap:.1f}s)")
         return selected
 
     def adjust_to_scene_boundaries(
@@ -246,6 +284,7 @@ class ClipExtractor:
         num_clips: Optional[int] = None,
         generate_thumbnails: bool = True,
         reencode: bool = False,
+        video_duration: Optional[float] = None,
     ) -> List[ExtractedClip]:
         """
         Extract clips from video.
@@ -257,6 +296,7 @@ class ClipExtractor:
             num_clips: Number of clips to extract (None = use config default)
             generate_thumbnails: Whether to generate thumbnails
             reencode: Whether to re-encode clips (slower but precise)
+            video_duration: Total video duration for diversity calculation
 
         Returns:
             List of ExtractedClip objects
@@ -266,8 +306,8 @@ class ClipExtractor:
         num_clips = num_clips or self.config.default_num_clips
 
         with LogTimer(logger, f"Extracting {num_clips} clips"):
-            # Select top clips
-            selected = self.select_clips(candidates, num_clips)
+            # Select top clips with adaptive diversity
+            selected = self.select_clips(candidates, num_clips, video_duration=video_duration)
 
             if not selected:
                 logger.warning("No clips to extract")