mapillary-tools 0.14.0a1__py3-none-any.whl → 0.14.0b1__py3-none-any.whl

mapillary_tools/ffmpeg.py

@@ -1,4 +1,5 @@
 # pyre-ignore-all-errors[5, 24]
+from __future__ import annotations
 
 import datetime
 import json
@@ -12,8 +13,7 @@ import typing as T
 from pathlib import Path
 
 LOG = logging.getLogger(__name__)
-FRAME_EXT = ".jpg"
-NA_STREAM_IDX = "NA"
+_MAX_STDERR_LENGTH = 2048
 
 
 class StreamTag(T.TypedDict):
@@ -30,19 +30,19 @@ class Stream(T.TypedDict):
     index: int
     tags: StreamTag
     width: int
+    r_frame_rate: str
+    avg_frame_rate: str
+    nb_frames: str
 
 
 class ProbeOutput(T.TypedDict):
-    streams: T.List[Stream]
+    streams: list[Stream]
 
 
 class FFmpegNotFoundError(Exception):
     pass
 
 
-_MAX_STDERR_LENGTH = 2048
-
-
 def _truncate_begin(s: str) -> str:
     if _MAX_STDERR_LENGTH < len(s):
         return "..." + s[-_MAX_STDERR_LENGTH:]
@@ -73,79 +73,43 @@ class FFmpegCalledProcessError(Exception):
 
 
 class FFMPEG:
+    FRAME_EXT = ".jpg"
+
     def __init__(
         self,
         ffmpeg_path: str = "ffmpeg",
         ffprobe_path: str = "ffprobe",
-        stderr: T.Optional[int] = None,
+        stderr: int | None = None,
     ) -> None:
         """
-        ffmpeg_path: path to ffmpeg binary
-        ffprobe_path: path to ffprobe binary
-        stderr: param passed to subprocess.run to control whether to capture stderr
+        Initialize FFMPEG wrapper with paths to ffmpeg and ffprobe binaries.
+
+        Args:
+            ffmpeg_path: Path to ffmpeg binary executable
+            ffprobe_path: Path to ffprobe binary executable
+            stderr: Parameter passed to subprocess.run to control stderr capture.
+                   Use subprocess.PIPE to capture stderr, None to inherit from parent
         """
         self.ffmpeg_path = ffmpeg_path
         self.ffprobe_path = ffprobe_path
         self.stderr = stderr
 
-    def _run_ffprobe_json(self, cmd: T.List[str]) -> T.Dict:
-        full_cmd: T.List[str] = [self.ffprobe_path, "-print_format", "json", *cmd]
-        LOG.info(f"Extracting video information: {' '.join(full_cmd)}")
-        try:
-            completed = subprocess.run(
-                full_cmd,
-                check=True,
-                stdout=subprocess.PIPE,
-                stderr=self.stderr,
-            )
-        except FileNotFoundError:
-            raise FFmpegNotFoundError(
-                f'The ffprobe command "{self.ffprobe_path}" not found'
-            )
-        except subprocess.CalledProcessError as ex:
-            raise FFmpegCalledProcessError(ex) from ex
-
-        try:
-            stdout = completed.stdout.decode("utf-8")
-        except UnicodeDecodeError:
-            raise RuntimeError(
-                f"Error decoding ffprobe output as unicode: {_truncate_end(str(completed.stdout))}"
-            )
-
-        try:
-            output = json.loads(stdout)
-        except json.JSONDecodeError:
-            raise RuntimeError(
-                f"Error JSON decoding ffprobe output: {_truncate_end(stdout)}"
-            )
-
-        # This check is for macOS:
-        # ffprobe -hide_banner -print_format json not_exists
-        # you will get exit code == 0 with the following stdout and stderr:
-        # {
-        # }
-        # not_exists: No such file or directory
-        if not output:
-            raise RuntimeError(
-                f"Empty JSON ffprobe output with STDERR: {_truncate_begin(str(completed.stderr))}"
-            )
+    def probe_format_and_streams(self, video_path: Path) -> ProbeOutput:
+        """
+        Probe video file to extract format and stream information using ffprobe.
 
-        return output
+        Args:
+            video_path: Path to the video file to probe
 
-    def _run_ffmpeg(self, cmd: T.List[str]) -> None:
-        full_cmd: T.List[str] = [self.ffmpeg_path, *cmd]
-        LOG.info(f"Extracting frames: {' '.join(full_cmd)}")
-        try:
-            subprocess.run(full_cmd, check=True, stderr=self.stderr)
-        except FileNotFoundError:
-            raise FFmpegNotFoundError(
-                f'The ffmpeg command "{self.ffmpeg_path}" not found'
-            )
-        except subprocess.CalledProcessError as ex:
-            raise FFmpegCalledProcessError(ex) from ex
+        Returns:
+            Dictionary containing streams and format information from ffprobe
 
-    def probe_format_and_streams(self, video_path: Path) -> ProbeOutput:
-        cmd: T.List[str] = [
+        Raises:
+            FFmpegNotFoundError: If ffprobe binary is not found
+            FFmpegCalledProcessError: If ffprobe command fails
+            RuntimeError: If output cannot be decoded or parsed as JSON
+        """
+        cmd: list[str] = [
             "-hide_banner",
             "-show_format",
             "-show_streams",
@@ -153,49 +117,79 @@ class FFMPEG:
         ]
         return T.cast(ProbeOutput, self._run_ffprobe_json(cmd))
 
-    def extract_frames(
+    def extract_frames_by_interval(
         self,
         video_path: Path,
         sample_dir: Path,
         sample_interval: float,
-        stream_idx: T.Optional[int] = None,
+        stream_specifier: int | str = "v",
     ) -> None:
         """
-        Extract frames by the sample interval from the specified video stream.
-
-        stream_idx: the stream_index specifier to a **video stream**. If it's None, defaults to "v". See http://ffmpeg.org/ffmpeg.html#Stream-specifiers-1
+        Extract frames from video at regular time intervals using fps filter.
+
+        Args:
+            video_path: Path to input video file
+            sample_dir: Directory where extracted frame images will be saved
+            sample_interval: Time interval between extracted frames in seconds
+            stream_specifier: Stream specifier to target specific stream(s).
+                              Can be an integer (stream index) or "v" (all video streams)
+                              See https://ffmpeg.org/ffmpeg.html#Stream-specifiers-1
+
+        Raises:
+            FFmpegNotFoundError: If ffmpeg binary is not found
+            FFmpegCalledProcessError: If ffmpeg command fails
         """
+        self._validate_stream_specifier(stream_specifier)
+
         sample_prefix = sample_dir.joinpath(video_path.stem)
-        if stream_idx is not None:
-            stream_selector = ["-map", f"0:{stream_idx}"]
-            ouput_template = f"{sample_prefix}_{stream_idx}_%06d{FRAME_EXT}"
-            stream_specifier = f"{stream_idx}"
-        else:
-            stream_selector = []
-            ouput_template = f"{sample_prefix}_{NA_STREAM_IDX}_%06d{FRAME_EXT}"
-            stream_specifier = "v"
-
-        cmd: T.List[str] = [
-            # global options should be specified first
-            *["-hide_banner", "-nostdin"],
-            # input 0
+        stream_selector = ["-map", f"0:{stream_specifier}"]
+        output_template = f"{sample_prefix}_{stream_specifier}_%06d{self.FRAME_EXT}"
+
+        cmd: list[str] = [
+            # Global options should be specified first
+            *["-hide_banner"],
+            # Input 0
             *["-i", str(video_path)],
-            # select stream
+            # Select stream
             *stream_selector,
-            # filter videos
+            # Filter videos
             *["-vf", f"fps=1/{sample_interval}"],
-            # video quality level (or the alias -q:v)
-            *[f"-qscale:{stream_specifier}", "2"],
+            # Video quality level (or the alias -q:v)
             # -q:v=1 is the best quality but larger image sizes
             # see https://stackoverflow.com/a/10234065
             # *["-qscale:v", "1", "-qmin", "1"],
-            # output
-            ouput_template,
+            *["-qscale:v", "2"],
+            # Output
+            output_template,
         ]
 
-        self._run_ffmpeg(cmd)
+        self.run_ffmpeg_non_interactive(cmd)
+
+    @classmethod
+    def generate_binary_search(cls, sorted_frame_indices: list[int]) -> str:
+        """
+        Generate a binary search expression for ffmpeg select filter.
+
+        Creates an optimized filter expression that uses binary search logic
+        to efficiently select specific frame numbers from a video stream.
+
+        Args:
+            sorted_frame_indices: List of frame numbers to select, must be sorted in ascending order
+
+        Returns:
+            FFmpeg filter expression string using binary search logic
+
+        Examples:
+            >>> FFMPEG.generate_binary_search([])
+            '0'
+            >>> FFMPEG.generate_binary_search([1])
+            'eq(n\\\\,1)'
+            >>> FFMPEG.generate_binary_search([1, 2])
+            'if(lt(n\\\\,2)\\\\,eq(n\\\\,1)\\\\,eq(n\\\\,2))'
+            >>> FFMPEG.generate_binary_search([1, 2, 3])
+            'if(lt(n\\\\,2)\\\\,eq(n\\\\,1)\\\\,if(lt(n\\\\,3)\\\\,eq(n\\\\,2)\\\\,eq(n\\\\,3)))'
+        """
 
-    def generate_binary_search(self, sorted_frame_indices: T.Sequence[int]) -> str:
         length = len(sorted_frame_indices)
 
         if length == 0:
@@ -204,39 +198,50 @@ class FFMPEG:
         if length == 1:
             return f"eq(n\\,{sorted_frame_indices[0]})"
 
-        middle = length // 2
-        return f"if(lt(n\\,{sorted_frame_indices[middle]})\\,{self.generate_binary_search(sorted_frame_indices[:middle])}\\,{self.generate_binary_search(sorted_frame_indices[middle:])})"
+        middle_idx = length // 2
+        left = cls.generate_binary_search(sorted_frame_indices[:middle_idx])
+        right = cls.generate_binary_search(sorted_frame_indices[middle_idx:])
+
+        return f"if(lt(n\\,{sorted_frame_indices[middle_idx]})\\,{left}\\,{right})"
 
     def extract_specified_frames(
         self,
         video_path: Path,
         sample_dir: Path,
-        frame_indices: T.Set[int],
-        stream_idx: T.Optional[int] = None,
+        frame_indices: set[int],
+        stream_specifier: int | str = "v",
     ) -> None:
         """
-        Extract specified frames from the specified video stream.
-
-        stream_idx: the stream_index specifier to a **video stream**. If it's None, defaults to "v". See http://ffmpeg.org/ffmpeg.html#Stream-specifiers-1
+        Extract specific frames from video by frame number using select filter.
+
+        Uses a binary search filter expression to efficiently select only the
+        specified frame numbers from the video stream.
+
+        Args:
+            video_path: Path to input video file
+            sample_dir: Directory where extracted frame images will be saved
+            frame_indices: Set of specific frame numbers to extract (0-based)
+            stream_specifier: Stream specifier to target specific stream(s).
+                              Can be an integer (stream index) or "v" (all video streams)
+                              See https://ffmpeg.org/ffmpeg.html#Stream-specifiers-1
+
+        Raises:
+            FFmpegNotFoundError: If ffmpeg binary is not found
+            FFmpegCalledProcessError: If ffmpeg command fails
+
+        Note:
+            Frame indices are 0-based but ffmpeg output files are numbered starting from 1.
+            Creates temporary filter script file on Windows to avoid command line length limits.
         """
 
+        self._validate_stream_specifier(stream_specifier)
+
         if not frame_indices:
             return
 
         sample_prefix = sample_dir.joinpath(video_path.stem)
-        if stream_idx is not None:
-            stream_selector = ["-map", f"0:{stream_idx}"]
-            ouput_template = f"{sample_prefix}_{stream_idx}_%06d{FRAME_EXT}"
-            stream_specifier = f"{stream_idx}"
-        else:
-            stream_selector = []
-            ouput_template = f"{sample_prefix}_{NA_STREAM_IDX}_%06d{FRAME_EXT}"
-            stream_specifier = "v"
-
-        # Write the select filter to a temp file because:
-        # The select filter could be large and
-        # the maximum command line length for the CreateProcess function is 32767 characters
-        # https://devblogs.microsoft.com/oldnewthing/20031210-00/?p=41553
+        stream_selector = ["-map", f"0:{stream_specifier}"]
+        output_template = f"{sample_prefix}_{stream_specifier}_%06d{self.FRAME_EXT}"
 
         eqs = self.generate_binary_search(sorted(frame_indices))
 
@@ -246,6 +251,10 @@ class FFMPEG:
         else:
             delete = True
 
+        # Write the select filter to a temp file because:
+        # The select filter could be large and
+        # the maximum command line length for the CreateProcess function is 32767 characters
+        # https://devblogs.microsoft.com/oldnewthing/20031210-00/?p=41553
         with tempfile.NamedTemporaryFile(mode="w+", delete=delete) as select_file:
             try:
                 select_file.write(f"select={eqs}")
@@ -253,14 +262,14 @@ class FFMPEG:
                 # If not close, error "The process cannot access the file because it is being used by another process"
                 if not delete:
                     select_file.close()
-                cmd: T.List[str] = [
-                    # global options should be specified first
-                    *["-hide_banner", "-nostdin"],
-                    # input 0
+                cmd: list[str] = [
+                    # Global options should be specified first
+                    *["-hide_banner"],
+                    # Input 0
                     *["-i", str(video_path)],
-                    # select stream
+                    # Select stream
                     *stream_selector,
-                    # filter videos
+                    # Filter videos
                     *[
                         *["-filter_script:v", select_file.name],
                         # Each frame is passed with its timestamp from the demuxer to the muxer
@@ -268,8 +277,8 @@ class FFMPEG:
                         # vsync is deprecated by fps_mode,
                         # but fps_mode is not avaliable on some older versions ;(
                         # *[f"-fps_mode:{stream_specifier}", "passthrough"],
-                        # Set the number of video frames to output
-                        *[f"-frames:{stream_specifier}", str(len(frame_indices))],
+                        # Set the number of video frames to output (this is an optimization to let ffmpeg stop early)
+                        *["-frames:v", str(len(frame_indices))],
                         # Disabled because it doesn't always name the sample images as expected
                         # For example "select(n\,1)" we expected the first sample to be IMG_001.JPG
                         # but it could be IMG_005.JPG
@@ -277,15 +286,15 @@ class FFMPEG:
                         # If set to 1, expand the filename with pts from pkt->pts. Default value is 0.
                         # *["-frame_pts", "1"],
                     ],
-                    # video quality level (or the alias -q:v)
-                    *[f"-qscale:{stream_specifier}", "2"],
+                    # Video quality level (or the alias -q:v)
                     # -q:v=1 is the best quality but larger image sizes
                     # see https://stackoverflow.com/a/10234065
                     # *["-qscale:v", "1", "-qmin", "1"],
+                    *["-qscale:v", "2"],
                     # output
-                    ouput_template,
+                    output_template,
                 ]
-                self._run_ffmpeg(cmd)
+                self.run_ffmpeg_non_interactive(cmd)
             finally:
                 if not delete:
                     try:
@@ -293,45 +302,286 @@ class FFMPEG:
                     except FileNotFoundError:
                         pass
 
+    @classmethod
+    def sort_selected_samples(
+        cls,
+        sample_dir: Path,
+        video_path: Path,
+        selected_stream_specifiers: list[int | str] | None = None,
+    ) -> list[tuple[int, list[Path | None]]]:
+        """
+        Group extracted frame samples by frame index across multiple streams.
+
+        Groups frames so that the Nth group contains all frames from the selected
+        streams at frame index N, allowing synchronized access to multi-stream frames.
+
+        Args:
+            sample_dir: Directory containing extracted frame files
+            video_path: Original video file path (used to match frame filenames)
+            selected_stream_specifiers: List of stream specifiers to include in output.
+                                       Can contain integers (stream indices) or "v" (all video streams).
+                                       If None, defaults to ["v"]
+
+        Returns:
+            List of tuples where each tuple contains:
+            - frame_idx (int): The frame index
+            - sample_paths (list[Path | None]): Paths to frame files from each selected stream,
+              or None if no frame exists for that stream at this index
+
+        Note:
+            Output is sorted by frame index in ascending order.
+        """
+        if selected_stream_specifiers is None:
+            selected_stream_specifiers = ["v"]
+
+        for stream_specifier in selected_stream_specifiers:
+            cls._validate_stream_specifier(stream_specifier)
+
+        stream_samples: dict[int, list[tuple[str, Path]]] = {}
+        for stream_specifier, frame_idx, sample_path in cls.iterate_samples(
+            sample_dir, video_path
+        ):
+            stream_samples.setdefault(frame_idx, []).append(
+                (str(stream_specifier), sample_path)
+            )
+
+        selected: list[tuple[int, list[Path | None]]] = []
+        for frame_idx in sorted(stream_samples.keys()):
+            indexed_by_specifier = {
+                specifier: sample_path
+                for specifier, sample_path in stream_samples[frame_idx]
+            }
+            selected_sample_paths = [
+                indexed_by_specifier.get(str(specifier))
+                for specifier in selected_stream_specifiers
+            ]
+            selected.append((frame_idx, selected_sample_paths))
+        return selected
+
+    @classmethod
+    def iterate_samples(
+        cls, sample_dir: Path, video_path: Path
+    ) -> T.Generator[tuple[str, int, Path], None, None]:
+        """
+        Iterate over all extracted frame samples in a directory.
+
+        Searches for frame files matching the expected naming pattern and yields
+        information about each frame including stream specifier, frame index, and file path.
+
+        Args:
+            sample_dir: Directory containing extracted frame files
+            video_path: Original video file path (used to match frame filenames)
+
+        Yields:
+            Tuple containing:
+            - stream_specifier (str): Stream specifier (number or "v")
+            - frame_idx (int): Frame index (0-based or 1-based depending on extraction method)
+            - sample_path (Path): Path to the frame image file
+
+        Note:
+            Expected filename pattern: {video_stem}_{stream_specifier}_{frame_idx:06d}.jpg
+            where stream_specifier can be a number or "v" for video streams.
+        """
+        sample_basename_pattern = re.compile(
+            rf"""
+            ^{re.escape(video_path.stem)}  # Match the video stem
+            _(?P<stream_specifier>\d+|v)   # Stream specifier can be a number or "v"
+            _(?P<frame_idx>\d+)$           # Frame index, can be 0-padded
+            """,
+            re.X,
+        )
+        for sample_path in sample_dir.iterdir():
+            result = cls._extract_stream_frame_idx(
+                sample_path.name, sample_basename_pattern
+            )
+            if result is not None:
+                stream_specifier, frame_idx = result
+                yield (stream_specifier, frame_idx, sample_path)
+
+    def run_ffmpeg_non_interactive(self, cmd: list[str]) -> None:
+        """
+        Execute ffmpeg command in non-interactive mode.
+
+        Runs ffmpeg with the given command arguments, automatically adding
+        the -nostdin flag to prevent interactive prompts.
+
+        Args:
+            cmd: List of command line arguments to pass to ffmpeg
+
+        Raises:
+            FFmpegNotFoundError: If ffmpeg binary is not found
+            FFmpegCalledProcessError: If ffmpeg command fails
+        """
+        full_cmd: list[str] = [self.ffmpeg_path, "-nostdin", *cmd]
+        LOG.info(f"Running ffmpeg: {' '.join(full_cmd)}")
+        try:
+            subprocess.run(full_cmd, check=True, stderr=self.stderr)
+        except FileNotFoundError:
+            raise FFmpegNotFoundError(
+                f'The ffmpeg command "{self.ffmpeg_path}" not found'
+            )
+        except subprocess.CalledProcessError as ex:
+            raise FFmpegCalledProcessError(ex) from ex
+
+    @classmethod
+    def _extract_stream_frame_idx(
+        cls, sample_basename: str, pattern: T.Pattern[str]
+    ) -> tuple[str, int] | None:
+        """
+        Extract stream specifier and frame index from sample basename
+
+        Returns:
+            If returning None, it means the basename does not match the pattern
+
+        Examples:
+            * basename GX010001_v_000000.jpg will extract ("v", 0)
+            * basename GX010001_1_000002.jpg will extract ("1", 2)
+        """
+        image_no_ext, ext = os.path.splitext(sample_basename)
+        if ext.lower() != cls.FRAME_EXT.lower():
+            return None
+
+        match = pattern.match(image_no_ext)
+        if not match:
+            return None
+
+        stream_specifier = match.group("stream_specifier")
+
+        # Convert 0-padded numbers to int
+        # e.g. 000000 -> 0
+        # e.g. 000001 -> 1
+        frame_idx_str = match.group("frame_idx")
+        frame_idx_str = frame_idx_str.lstrip("0") or "0"
+
+        try:
+            frame_idx = int(frame_idx_str)
+        except ValueError:
+            return None
+
+        return stream_specifier, frame_idx
+
+    def _run_ffprobe_json(self, cmd: list[str]) -> dict:
+        full_cmd: list[str] = [self.ffprobe_path, "-print_format", "json", *cmd]
+        LOG.info(f"Extracting video information: {' '.join(full_cmd)}")
+        try:
+            completed = subprocess.run(
+                full_cmd, check=True, stdout=subprocess.PIPE, stderr=self.stderr
+            )
+        except FileNotFoundError:
+            raise FFmpegNotFoundError(
+                f'The ffprobe command "{self.ffprobe_path}" not found'
+            )
+        except subprocess.CalledProcessError as ex:
+            raise FFmpegCalledProcessError(ex) from ex
+
+        try:
+            stdout = completed.stdout.decode("utf-8")
+        except UnicodeDecodeError:
+            raise RuntimeError(
+                f"Error decoding ffprobe output as unicode: {_truncate_end(str(completed.stdout))}"
+            )
+
+        try:
+            output = json.loads(stdout)
+        except json.JSONDecodeError:
+            raise RuntimeError(
+                f"Error JSON decoding ffprobe output: {_truncate_end(stdout)}"
+            )
+
+        # This check is for macOS:
+        # ffprobe -hide_banner -print_format json not_exists
+        # you will get exit code == 0 with the following stdout and stderr:
+        # {
+        # }
+        # not_exists: No such file or directory
+        if not output:
+            raise RuntimeError(
+                f"Empty JSON ffprobe output with STDERR: {_truncate_begin(str(completed.stderr))}"
+            )
+
+        return output
+
+    @classmethod
+    def _validate_stream_specifier(cls, stream_specifier: int | str) -> None:
+        if isinstance(stream_specifier, str):
+            if stream_specifier in ["v"]:
+                pass
+            else:
+                try:
+                    int(stream_specifier)
+                except ValueError:
+                    raise ValueError(f"Invalid stream specifier: {stream_specifier}")
+
 
 class Probe:
-    probe: ProbeOutput
+    probe_output: ProbeOutput
 
-    def __init__(self, probe: ProbeOutput) -> None:
-        self.probe = probe
+    def __init__(self, probe_output: ProbeOutput) -> None:
+        """
+        Initialize Probe with ffprobe output data.
 
-    def probe_video_start_time(self) -> T.Optional[datetime.datetime]:
+        Args:
+            probe_output: Dictionary containing streams and format information from ffprobe
         """
-        Find video start time of the given video.
-        It searches video creation time and duration in video streams first and then the other streams.
-        Once found, return stream creation time - stream duration as the video start time.
+        self.probe_output = probe_output
+
+    def probe_video_start_time(self) -> datetime.datetime | None:
         """
-        streams = self.probe.get("streams", [])
+        Determine the start time of the video by analyzing stream metadata.
+
+        Searches for creation time and duration information in video streams first,
+        then falls back to other stream types. Calculates start time as:
+        creation_time - duration
 
-        # search start time from video streams
+        Returns:
+            Video start time as datetime object, or None if cannot be determined
+
+        Note:
+            Prioritizes video streams with highest resolution when multiple exist.
+        """
+        streams = self.probe_output.get("streams", [])
+
+        # Search start time from video streams
         video_streams = self.probe_video_streams()
         video_streams.sort(
             key=lambda s: s.get("width", 0) * s.get("height", 0), reverse=True
         )
         for stream in video_streams:
-            start_time = extract_stream_start_time(stream)
+            start_time = self.extract_stream_start_time(stream)
             if start_time is not None:
                 return start_time
 
-        # search start time from the other streams
+        # Search start time from the other streams
         for stream in streams:
             if stream.get("codec_type") != "video":
-                start_time = extract_stream_start_time(stream)
+                start_time = self.extract_stream_start_time(stream)
                 if start_time is not None:
                     return start_time
 
         return None
 
-    def probe_video_streams(self) -> T.List[Stream]:
-        streams = self.probe.get("streams", [])
+    def probe_video_streams(self) -> list[Stream]:
+        """
+        Extract all video streams from the probe output.
+
+        Returns:
+            List of video stream dictionaries containing metadata like codec,
+            dimensions, frame rate, etc.
+        """
+        streams = self.probe_output.get("streams", [])
         return [stream for stream in streams if stream.get("codec_type") == "video"]
 
-    def probe_video_with_max_resolution(self) -> T.Optional[Stream]:
+    def probe_video_with_max_resolution(self) -> Stream | None:
+        """
+        Find the video stream with the highest resolution.
+
+        Sorts all video streams by width × height and returns the one with
+        the largest resolution.
+
+        Returns:
+            Stream dictionary for the highest resolution video stream,
+            or None if no video streams exist
+        """
         video_streams = self.probe_video_streams()
         video_streams.sort(
             key=lambda s: s.get("width", 0) * s.get("height", 0), reverse=True
@@ -340,112 +590,37 @@ class Probe:
             return None
         return video_streams[0]
 
+    @classmethod
+    def extract_stream_start_time(cls, stream: Stream) -> datetime.datetime | None:
+        """
+        Calculate the start time of a specific stream.
 
-def extract_stream_start_time(stream: Stream) -> T.Optional[datetime.datetime]:
-    """
-    Find the start time of the given stream.
-    Start time is the creation time of the stream minus the duration of the stream.
-    """
-    duration_str = stream.get("duration")
-    LOG.debug("Extracted video duration: %s", duration_str)
-    if duration_str is None:
-        return None
-    duration = float(duration_str)
+        Determines start time by subtracting stream duration from creation time:
+        start_time = creation_time - duration
 
-    creation_time_str = stream.get("tags", {}).get("creation_time")
-    LOG.debug("Extracted video creation time: %s", creation_time_str)
-    if creation_time_str is None:
-        return None
-    try:
-        creation_time = datetime.datetime.fromisoformat(creation_time_str)
-    except ValueError:
-        creation_time = datetime.datetime.strptime(
-            creation_time_str, "%Y-%m-%dT%H:%M:%S.%f%z"
-        )
-    return creation_time - datetime.timedelta(seconds=duration)
-
-
-def _extract_stream_frame_idx(
-    sample_basename: str,
-    sample_basename_pattern: T.Pattern[str],
-) -> T.Optional[T.Tuple[T.Optional[int], int]]:
-    """
-    extract stream id and frame index from sample basename
-    e.g. basename GX010001_NA_000000.jpg will extract (None, 0)
-    e.g. basename GX010001_1_000002.jpg will extract (1, 2)
-    If returning None, it means the basename does not match the pattern
-    """
-    image_no_ext, ext = os.path.splitext(sample_basename)
-    if ext.lower() != FRAME_EXT.lower():
-        return None
+        Args:
+            stream: Stream dictionary containing metadata including tags and duration
 
-    match = sample_basename_pattern.match(image_no_ext)
-    if not match:
-        return None
+        Returns:
+            Stream start time as datetime object, or None if required metadata is missing
 
-    g1 = match.group("stream_idx")
-    try:
-        if g1 == NA_STREAM_IDX:
-            stream_idx = None
-        else:
-            stream_idx = int(g1)
-    except ValueError:
-        return None
-
-    # convert 0-padded numbers to int
-    # e.g. 000000 -> 0
-    # e.g. 000001 -> 1
-    g2 = match.group("frame_idx")
-    g2 = g2.lstrip("0") or "0"
-
-    try:
-        frame_idx = int(g2)
-    except ValueError:
-        return None
+        Note:
+            Handles multiple datetime formats including ISO format and custom patterns.
+        """
+        duration_str = stream.get("duration")
+        LOG.debug("Extracted video duration: %s", duration_str)
+        if duration_str is None:
+            return None
+        duration = float(duration_str)
 
-    return stream_idx, frame_idx
-
-
-def iterate_samples(
-    sample_dir: Path, video_path: Path
-) -> T.Generator[T.Tuple[T.Optional[int], int, Path], None, None]:
-    """
-    Search all samples in the sample_dir,
-    and return a generator of the tuple: (stream ID, frame index, sample path).
-    The frame index could be 0-based or 1-based depending on how it's sampled.
-    """
-    sample_basename_pattern = re.compile(
-        rf"^{re.escape(video_path.stem)}_(?P<stream_idx>\d+|{re.escape(NA_STREAM_IDX)})_(?P<frame_idx>\d+)$"
-    )
-    for sample_path in sample_dir.iterdir():
-        stream_frame_idx = _extract_stream_frame_idx(
-            sample_path.name,
-            sample_basename_pattern,
-        )
-        if stream_frame_idx is not None:
-            stream_idx, frame_idx = stream_frame_idx
-            yield (stream_idx, frame_idx, sample_path)
-
-
-def sort_selected_samples(
-    sample_dir: Path, video_path: Path, selected_stream_indices: T.List[T.Optional[int]]
-) -> T.List[T.Tuple[int, T.List[T.Optional[Path]]]]:
-    """
-    Group frames by frame index, so that
-    the Nth group contains all the frames from the selected streams at frame index N.
-    """
-    stream_samples: T.Dict[int, T.List[T.Tuple[T.Optional[int], Path]]] = {}
-    for stream_idx, frame_idx, sample_path in iterate_samples(sample_dir, video_path):
-        stream_samples.setdefault(frame_idx, []).append((stream_idx, sample_path))
-
-    selected: T.List[T.Tuple[int, T.List[T.Optional[Path]]]] = []
-    for frame_idx in sorted(stream_samples.keys()):
-        indexed = {
-            stream_idx: sample_path
-            for stream_idx, sample_path in stream_samples[frame_idx]
-        }
-        selected_sample_paths = [
-            indexed.get(stream_idx) for stream_idx in selected_stream_indices
-        ]
-        selected.append((frame_idx, selected_sample_paths))
-    return selected
+        creation_time_str = stream.get("tags", {}).get("creation_time")
+        LOG.debug("Extracted video creation time: %s", creation_time_str)
+        if creation_time_str is None:
+            return None
+        try:
+            creation_time = datetime.datetime.fromisoformat(creation_time_str)
+        except ValueError:
+            creation_time = datetime.datetime.strptime(
+                creation_time_str, "%Y-%m-%dT%H:%M:%S.%f%z"
+            )
+        return creation_time - datetime.timedelta(seconds=duration)