Skip to content

v0.1.2-2 (pre-release) — training recommendations, ONNX/TensorRT export, track parity

Pre-release
Pre-release

Choose a tag to compare

View all tags
@gitttt-1234 gitttt-1234 released this 31 Aug 22:46
· 16 commits to main since this release
v0.1.2-2
5cff4f5
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Second pre-release of the 0.1.2 line — 40 commits since v0.1.2-1.

Training

  • Auto max_stride/crop_size recommendations plus GPU and cache memory estimates (#334)
  • LR scheduler, checkpoint retention, and eval-metric controls (#342)
  • Multi-GPU strategy dropdown (trainer_strategy) (#335)
  • Resume training and Fine-tune modes wired up (#305)
  • OHKM hard/easy ratio and loss scale (#311)
  • WandB offline mode + API key field (#312)
  • Compact Full Configuration dialog chrome: flat tabs, saved indicator, reset & Done (#315)
  • Radio-group training mode, auto color mode, tooltips, stale-footer fix (#310)

Inference & export

  • In-app ONNX/TensorRT model exporter with on-demand install (#338)
  • ONNX/TensorRT runtime selection via --runtime (#337)
  • Track-only workflow; untracked instances greyed out (#323)
  • Full sleap-nn tracking and post-processing filter params exposed (#309)
  • "Existing predictions" is honored, so re-inference stops stacking duplicates (#303)

Labeling & canvas

  • Track propagation is now O(scope) instead of O(project), with legacy PyQt parity for track assignment (#336)
  • Live labeling hints for common novice pitfalls (#352)
  • Readable track labels: halo, de-collision, sizing, score toggle (#318)
  • Add-vs-replace on Generate, plus shuffle / add-labeled / remove-unlabeled tools (#332)
  • Node dragging no longer freezes the GUI (#331)
  • Sidebar tables stay scrollable with working sticky headers (#347)
  • OS-correct shortcut labels, multi-panel default, labels.vNNN.slp auto-versioning (#302)
  • Interactive getting-started walkthrough (#294)
  • UI polish: accordion side-tabs, toast Clear-all, rounded macOS icon (#317)

Video & project I/O

  • Add a video from a URL (#322)
  • Drag-and-drop video dropzone in the Videos panel and New Project (#319)
  • Drop an .slp onto a loaded project to open it in a new window (#330)
  • "Open in SLEAP" browser deep link — stream a remote .slp from ?open=<url> (#284)
  • In-app .avi/.wmv labeling via the sleap-io.js AviVideoBackend (#306)
  • Desktop legacy-codec transcode fallback: Xvid/WMV/MPEG → H.264 (#307)
  • Import videos as grayscale, matching the legacy Qt GUI (#308)
  • Cross-video suggestion navigation and all-videos default (#333)

Desktop, updater & CI

  • Channel-switch downgrade support and a rustls crash fix (#353)
  • Stable/latest/dev auto-update channels with an Environment badge (#314)
  • macOS notarization for dev builds, split updater artifact (#350)
  • Renamed macOS build artifacts and Environment channel fixes (#351)
  • NSIS-only Windows dev builds, channel-aware install scripts (#348)
  • confirm/alert/prompt routed through in-app dialogs, fixing Tauri data loss (#321)
  • No more false "didn't close properly" recovery prompt (#320)
  • Web deploys now run after the desktop build succeeds, so a release is atomic — if a platform's installer fails to build, app.sleap.ai doesn't start serving that version either. Each deployed path is also stamped with the version its desktop counterpart reports, instead of the stale committed 0.1.0. This also makes app.sleap.ai/latest/ work for the first time: its gate required an asset that only exists after the build, so it could never pass when checked at publish time (#354)
  • Vendored ffmpeg/ffprobe sidecars so the desktop release build stops failing (#313)
Assets 16
Loading