The 2024-04-08 total solar eclipse, from the Port of Cleveland on Lake Erie,
processed by this pipeline.
2299 frames, 76.6 seconds at 30 fps, all 22 captures, from ~449 GB of SER.
ZWO ASI585MC on a TS-Optics 70 mm APO with a 0.8x reducer — 350 mm at f/5,
1.72 arcsec per sensor pixel — on a Sky-Watcher EQ6-R Pro.
Produced by one command, with no arguments beyond the data directory:
python -m ecl.run <your captures>
Nothing here was hand-tuned. The output window (1180×880 plane px), the drizzle
factor, the worker count, the exposure normalisation, the frames kept and
dropped, the segments in the GIF and the feature every zoom panel follows were
all decided from the data. That is the point of the release: it is what the code
does, not what its author could coax out of it.
| asset | ||
|---|---|---|
eclipse-2024-04-08-2360x1760.mp4 |
93 MB | full resolution, CRF 17 |
eclipse-2024-04-08-1080x806.mp4 |
12 MB | sized so Instagram does not re-encode it |
eclipse-2024-04-08-960x716-preview.mp4 |
2 MB | quick look |
eclipse-2024-04-08-preview.gif |
4 MB | the animated preview, four segments the pipeline chose |
The Sun is held still by a fitted disc track rather than by frame correlation,
because correlation locks onto the brightest thing in frame and during the
partial phases that is the crescent — whose centroid is not the centre of the
disc it was cut from. docs/NOTES.md has the measurements behind that and every
other decision here.
These are release assets rather than files in the repository: the full-resolution
cut is over GitHub's 100 MiB per-file limit for git, and a video committed to git
would sit in every clone forever.
Supersedes an earlier upload of a hand-graded cut (2228 frames, 1200×900,
framed with --window 1120x840 --drop-padded --require-disc). The two are
photometrically identical where they overlap — holding the config fixed and
re-rendering gives byte-identical frames — they differ only in framing and in
which partial-phase frames were kept.