From e237073512de80ec5b6f2974606f0ba7ad21671e Mon Sep 17 00:00:00 2001
From: Dominik Winecki <dominikwinecki@gmail.com>
Date: Thu, 8 Jan 2026 20:00:48 +0000
Subject: [PATCH] Add support for VideoCapture.get(CAP_PROP_POS_MSEC)

---
 snake-pit/test_cv2.py                  | 45 ++++++++++++++++++++++++++
 vidformer-py/vidformer/cv2/__init__.py | 11 +++++--
 2 files changed, 54 insertions(+), 2 deletions(-)

diff --git a/snake-pit/test_cv2.py b/snake-pit/test_cv2.py
index 99ccdec6..f1c352f6 100644
--- a/snake-pit/test_cv2.py
+++ b/snake-pit/test_cv2.py
@@ -752,6 +752,51 @@ def test_seek_vf():
     seek(vf_cv2)
 
 
+def get_pos_msec(cv2):
+    # test getting position in milliseconds
+    cap = cv2.VideoCapture(TEST_VID_PATH)
+    assert cap.isOpened()
+
+    fps = cap.get(cv2.CAP_PROP_FPS)
+
+    # At start, should be at position 0
+    pos_msec = cap.get(cv2.CAP_PROP_POS_MSEC)
+    assert pos_msec == 0.0
+
+    # Seek to frame 24 (1 second at 24fps)
+    cap.set(cv2.CAP_PROP_POS_FRAMES, 24)
+    pos_msec = cap.get(cv2.CAP_PROP_POS_MSEC)
+    expected_msec = 24 * (1000 / fps)
+    assert abs(pos_msec - expected_msec) < 1.0  # within 1ms
+
+    # Seek to 5000ms using set, then verify with get
+    cap.set(cv2.CAP_PROP_POS_MSEC, 5000)
+    pos_msec = cap.get(cv2.CAP_PROP_POS_MSEC)
+    assert abs(pos_msec - 5000) < 2 * (1000 / fps)  # within two frame durations
+
+    # Seek to frame 1000
+    cap.set(cv2.CAP_PROP_POS_FRAMES, 1000)
+    pos_msec = cap.get(cv2.CAP_PROP_POS_MSEC)
+    expected_msec = 1000 * (1000 / fps)
+    assert abs(pos_msec - expected_msec) < 2 * (
+        1000 / fps
+    )  # within two frame durations
+
+    # Read a frame and verify position advances
+    pos_before = cap.get(cv2.CAP_PROP_POS_MSEC)
+    ret, frame = cap.read()
+    assert ret
+    pos_after = cap.get(cv2.CAP_PROP_POS_MSEC)
+    assert pos_after > pos_before
+
+    cap.release()
+
+
+# We can not run this with ocv_cv2 since OpenCV's get CAP_PROP_POS_MSEC is quite inaccurate
+def test_get_pos_msec_vf():
+    get_pos_msec(vf_cv2)
+
+
 def test_getFontScaleFromHeight():
     import cv2 as ocv_cv2
 
diff --git a/vidformer-py/vidformer/cv2/__init__.py b/vidformer-py/vidformer/cv2/__init__.py
index 79c1fb8e..72f1b28d 100644
--- a/vidformer-py/vidformer/cv2/__init__.py
+++ b/vidformer-py/vidformer/cv2/__init__.py
@@ -74,7 +74,6 @@ def _ts_to_fps(timestamps):
     return float(fps)
 
 
-
 def _fps_to_ts(fps, n_frames):
     assert type(fps) is int
     return [Fraction(i, fps) for i in range(n_frames)]
@@ -363,6 +362,14 @@ def get(self, prop):
             return len(self._source)
         elif prop == CAP_PROP_POS_FRAMES:
             return self._next_frame_idx
+        elif prop == CAP_PROP_POS_MSEC:
+            ts = self._source.ts()
+            if self._next_frame_idx >= len(ts):
+                # Past the end, return the last timestamp
+                if len(ts) > 0:
+                    return float(ts[-1] * 1000)
+                return 0.0
+            return float(ts[self._next_frame_idx] * 1000)
 
         raise Exception(f"Unknown property {prop}")
 
@@ -439,7 +446,7 @@ def __init__(
                 # Round to nearest integer fps
                 self._f_time = Fraction(1, int(round(fps)))
         else:
-            raise Exception("fps must be an integer or a Fraction")
+            raise Exception("fps must be an integer, float, or Fraction")
 
         assert isinstance(size, tuple) or isinstance(size, list)
         assert len(size) == 2