DashStack

DashStack stacks timestamp-matched dashcam clips vertically (front on top, rear on bottom) and concatenates them chronologically. It auto-detects the fastest encoder on your system — CUDA+NVENC GPU pipeline, multi-core CPU, or hardware encoder.

No external Python dependencies. Just ffmpeg.

Requirements

• Python 3.9+
• ffmpeg and ffprobe on PATH
• rsync on PATH (for upload/download subcommands)

Installation

pip install .

Or editable for development:

pip install -e .

Quick start

dashstack videos                  # split mode: one _FR file per continuous run
dashstack videos --dry-run        # preview the plan without encoding
dashstack videos --limit 3        # quick test with first 3 pairs
dashstack videos --clean          # encode, then delete source files covered by output

How it works

DashStack scans a directory for files matching *YYYYMMDD_HHMMSS_F.mp4 / *_R.mp4, pairs them by timestamp, and stacks each pair into a combined front+rear video.

Split mode (default)

When no --output is given, DashStack detects gaps between clips (default >5s) and produces one _FR output file per continuous run, written to the input directory:

REC_20260313_135042_F.MP4  ┐
REC_20260313_135042_R.MP4  ┤→ REC_20260313_135042_20260313_135242_FR.MP4
REC_20260313_135142_F.MP4  ┤
REC_20260313_135142_R.MP4  ┘

The end timestamp in _FR filenames reflects the actual end time of the video (start of last segment + its duration), not just the last segment's start.

On re-runs, existing _FR files are detected and reused — source clips already covered by an _FR file are skipped, so only new pairs get encoded. Existing _FR filenames are automatically corrected if their end timestamp is wrong or missing.

Event clips (EVT_*) typically include a pre-trigger buffer, so their filename timestamp marks the trigger moment, not the start of the file. When an EVT clip is followed by a regular REC_* clip, DashStack infers the EVT's real start time from the next clip's start (since the camera records continuously) and trims the prior REC_* accordingly — so the EVT plays in full, with no duplicated footage and no time jumps at the seam.

Single-file mode

Combine everything into one file:

dashstack videos --output combined.mp4

Cleanup (--clean)

The --clean flag deletes files that are fully covered by _FR output:

• Source _F/_R clips whose timestamps fall within an _FR range
• Older _FR files that are fully subsumed by a newer, larger _FR file

This means you can run dashstack videos --clean incrementally — as small _FR files get merged with new clips into bigger ones, the old _FR files are cleaned up too.

Subcommands

upload

Transfer files to a remote destination via rsync over SSH, with an overall progress bar.

dashstack upload video1_FR.MP4 video2_FR.MP4 user@host:/path/
dashstack upload --delete *.MP4 user@host:/path/   # delete local files after upload
dashstack upload --dry-run *.MP4 user@host:/path/   # preview without transferring

download

Pull files from a remote source via rsync over SSH.

dashstack download user@host:/path/*.MP4 ./videos/
dashstack download --delete user@host:/path/*.MP4   # delete remote files after download
dashstack download --dry-run user@host:/path/*.MP4   # preview without transferring

dedup

Find and remove duplicate portions in already-merged _FR videos. This handles videos that were merged before overlap trimming was added, or from any other source with repeated footage at clip boundaries.

Detection works by finding abrupt frame changes (clip boundaries) and verifying that the footage after the cut matches footage from a few seconds before it.

dashstack dedup video_FR.MP4                 # fix in-place
dashstack dedup --dry-run video_FR.MP4       # show duplicates without modifying
dashstack dedup --max-overlap 30 *.MP4       # search for overlaps up to 30s (default)

Stacking options

Flag	Description
input_dir	Source directory (default: .)
--output PATH	Single combined output file (omit for split mode)
--gap-threshold N	Seconds of gap to split runs (default: 5.0)
--clean	Delete source files and superseded _FR files covered by output
--dry-run	Preview the plan without encoding
--limit N	Process only the first N pairs
--pipeline auto|single-pass|segment	Processing strategy (default: auto)
--target-width N	Panel width for each camera stream
--video-codec NAME	Encoder (default: auto)
--video-bitrate RATE	Bitrate for HW codecs (default: 16M)
--crf N	Quality for libx264 mode (default: 20)
--preset NAME	Speed/quality for libx264 (default: veryfast)
--audio-source front|rear|none	Audio track to keep (default: front)
--missing skip|error	Unmatched timestamp handling (default: skip)
--workers N	Parallel ffmpeg workers (default: half of CPU cores, max 4)
--hwaccel METHOD	HW decoding: auto, cuda, vaapi, none, etc.
--work-dir PATH	Temp directory (default: .dashstack_work)
--keep-temp	Retain intermediate segment files
--overwrite	Overwrite existing output without prompting

Filename pattern

Clips must contain a YYYYMMDD_HHMMSS timestamp followed by _F or _R before the extension. The prefix is flexible:

REC_20260312_105549_F.MP4
REC_20260312_105549_R.MP4

Unmatched timestamps (only F or only R) are skipped by default.