[bugfix] Ensure ThresholdDetector produces deterministic boundary rounding

Author: Breakthrough

scenedetect/detectors/threshold_detector.py

@@ -145,11 +145,17 @@ def process_frame(
                 # Only add the scene if min_scene_len frames have passed.
                 if (timecode - self.last_scene_cut) >= self.min_scene_len:
                     # Just faded into a new scene, compute timecode for the scene
-                    # split based on the fade bias.
+                    # split based on the fade bias. Use frame-number arithmetic so the
+                    # result is identical across backends — float seconds + framerate
+                    # multiplication can land on a .5 rounding boundary and tip the
+                    # frame number by 1 between PyAV (sub-microsecond PTS) and OpenCV
+                    # (millisecond-truncated CAP_PROP_POS_MSEC).
                     f_out = self.last_fade["frame"]
-                    duration = (timecode - f_out).seconds
-                    split_seconds = f_out.seconds + (duration * (1.0 + self.fade_bias)) / 2.0
-                    cuts.append(FrameTimecode(split_seconds, fps=timecode))
+                    duration_frames = timecode.frame_num - f_out.frame_num
+                    split_frame_num = f_out.frame_num + round(
+                        duration_frames * (1.0 + self.fade_bias) / 2.0
+                    )
+                    cuts.append(FrameTimecode(split_frame_num, fps=timecode))
                     self.last_scene_cut = timecode
                 self.last_fade["type"] = "in"
                 self.last_fade["frame"] = timecode

tests/test_detectors.py

@@ -139,7 +139,7 @@ def get_fade_in_out_test_cases():
                 detector=ThresholdDetector(),
                 start_time=0,
                 end_time=500,
-                scene_boundaries=[0, 15, 198, 376],
+                scene_boundaries=[0, 15, 198, 377],
             ),
             id="threshold_testvideo_default",
         ),
@@ -177,7 +177,7 @@ def get_fade_in_out_test_cases():
                 ),
                 start_time=0,
                 end_time=250,
-                scene_boundaries=[0, 42, 125, 209],
+                scene_boundaries=[0, 42, 126, 209],
             ),
             id="threshold_fades_ceil",
         ),

website/pages/changelog.md

@@ -681,6 +681,7 @@ Although there have been minimal changes to most API examples, there are several
 - [feature] `save-edl` accepts a new `--start-timecode`/`-s` flag (SMPTE `HH:MM:SS:FF` or 8-digit `HHMMSSFF`) to stamp every event with a custom start timecode so generated EDLs align with the source media's on-screen timecode [#515](https://github.com/Breakthrough/PySceneDetect/issues/515)
 - [bugfix] Fix floating-point precision error in `save-otio` output where frame values near integer boundaries (e.g. `90.00000000000001`) were serialized with spurious precision
 - [bugfix] Add mitigation for transient `OSError` in the MoviePy backend as it is susceptible to subprocess pipe races on slow or heavily loaded systems [#496](https://github.com/Breakthrough/PySceneDetect/issues/496)
+- [bugfix] `detect-threshold` cut frame numbers are now backend-deterministic; previously the cut could differ by 1 frame between PyAV and OpenCV when the fade midpoint landed on a `.5` rounding boundary (PyAV uses sub-microsecond PTS, OpenCV uses millisecond-truncated `CAP_PROP_POS_MSEC`)
 - [refactor] Remove deprecated `-d`/`--min-delta-hsv` option from `detect-adaptive` command
 
 ### API Changes