Skip to content

v0.8.0 — Image timing, reviewed masks and continuous comparisons

Pre-release
Pre-release

Choose a tag to compare

View all tags
@Oceanswave Oceanswave released this 13 Sep 03:03
· 1 commit to main since this release
v0.8.0
77aeb2b
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

A1 Stitcher 0.8 adds image-based timing, reviewed obstruction proposals, multiband seams and continuous reference comparisons.

  • Compare moving shots: benchmark produces six fixed views and per-frame measurements over a contiguous range, with explicit source mapping and tracking coverage.
  • Fit image timing: sync-calibrate refines gyro offset and row readout. Fits must pass temporal holdouts; incompatible camera modes, gyro profiles and row settings are rejected.
  • Review camera masks: mask-propose creates native overlays; mask-approve enables reviewed exclusions. Forced replacements check alternate-lens clipping/contrast and fail when no usable view exists.
  • Blend broader brightness differences: opt-in --seam multiband preserves native detail, bounds gain changes and tapers corrections inside valid overlap.

The README and agent skill describe every new option and the importance of source/archive differences. Native 8K, 16-bit processing, 10-bit HEVC and automatic Metal remain the finishing defaults. Experimental timing, masks and multiband remain explicit choices.

Final matched 20-second comparisons reduced the p95 motion diagnostic by about 36% on the fitting interval and 24% on a separate recording. These are image-track measurements, not a universal image-quality score.

Validation includes 342 passing local tests, 66 actual Metal cases, Rockybot Python 3.12/3.13 CI, an outside-checkout 8K/10-bit wheel proof with GPX, and longer source-matched Studio comparisons. Measured results and limits separate motion, geometry, brightness, technical verification and sampled visual review.

Alpha: Studio remains smoother on the tested intervals. Complete moving-propeller segmentation, severe occlusion, flare and broader camera qualification remain open. Keep INSV originals. No footage, per-camera profiles, vendor binaries or models are distributed. The project has no vendor runtime dependency.

Install the attached wheel with Python 3.12+ and FFmpeg, or use the tagged GitHub source. No PyPI publication is implied.

Assets 4
Loading