Skip to content

feat(core/desktop): single-layer auto-tune loop (#89) - #90

Merged
loootte merged 1 commit into
mainfrom
feature/89-layer-tune-loop
Jul 31, 2026
Merged

loootte merged 1 commit into
mainfrom
feature/89-layer-tune-loop

Conversation

@loootte

@loootte loootte commented Jul 31, 2026

Copy link
Copy Markdown
Owner

Summary

Closes the single-layer fit loop for #89 (L3 first):

?? ? ??? GT ? ?? ? ???? ? ?? ? ???

Core

  • configs/tune/ default + search spaces (L3/L4)
  • app/tuning/: params store, match_boxes / layer_loss, tune_layer random/grid search with seed & budget
  • L3/L4 read thresholds from params (no silent magic-only path)
  • API: /v1/evaluation/tune-layer[/upload], params/apply|reset|GET
  • CLI: scripts/tune_layer.py

Desktop

Tests

  • test_tuning_layer.py (match, loss, reproducible search, param effect)
  • Existing L3/eval suites green

Test plan

  • pytest tuning + l3 + evaluation
  • tsc --noEmit
  • UI: save edit GT ? auto-tune L3 ? apply ? L2 re-run

Closes #89

Parameterize L3/L4 from configs/tune YAML, random/grid search minimizing
layer IoU loss against edit-as-GT, apply best to runtime store, CLI and
API endpoints, and desktop controls to tune / apply / re-run from L2.
@loootte
loootte merged commit 95bdb11 into main Jul 31, 2026
3 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

feat(core/desktop): 单层参数自动调优循环(识别 → 手动框 GT → 评估 → 调参 → 再识别)

1 participant