Conversation
Builds the human half of the dataset-refinement loop on top of the TagSuggestion model: review near-neighbour disagreements one tag at a time and write fixes back to the labels. Backend (services/tag_suggestion_service.py, routes/tag_suggestions.py): - list/summary/accept/dismiss/reopen plus fix-twin and swap, all undoable (reopen reverses the exact label change: ACCEPTED/TWIN_FIXED/SWAPPED). - Surfaces each pair's picture format (for full-res rendering) and the tagger's per-image confidence. - bulk-accept: decide each pair from the two images' tagger confidence (the near-neighbour link only selects the pair), place it in one of four corners (both / left-only / right-only / neither) and auto-apply the confident ones above a threshold; dry-run returns a sample of the least-confident would-be resolutions to preview, and bulk-reopen batch-undoes. Merges (db_models/tag.py): DEFAULT_TAG_MERGES mirrors pixltagger's tag_remap as an editable constant; the scanner treats a child tag (e.g. "extra digit") as the parent for voting + "missing" eligibility but uses the literal tag for "remove" eligibility. Scanner also gains --tag-remap and --replace. Frontend (ReviewFixesOverlay.vue + useReviewFixesStore.js): a Toolbar- launched overlay framed as "the left is flagged - but is it?", with the 2x2 which-has-it answer (incl. swap), per-image tagger confidence, click- to-zoom/pan, keyboard review, undo, and the bulk-resolve bar with a preview modal. vite.config.js: restore outDir to ../pixlstash/frontend/dist (the in-repo package's served dist) with a comment to prevent the doubled-path mistake.
Tagger-flywheel progress checkpoint, framed by the new design notes. PixlStash as system of record (stats): - TaggerRun model + migration 0059 (chained after 0058_split_caption_sidecars) - POST/GET /tagger-runs: upsert by run name, stores every run incl. rejected, keeps the full report.json payload for the trend/stats panel - Registered in server.py; exported from db_models; 2 API tests Review-queue refinements (on top of 23bdf73): - 4-corner per-image decision in the overlay/store (drop the old direction filter) - pair dedupe in the shared near-neighbour kernel; in-app tag scan; autocomplete Design notes (docs/reviews/): - pixlstash-model-platform-plan: reframes PixlStash as a diffusion dataset & training platform (tagger + LoRA share a GPU-exclusive TrainingJob run via the task manager); records where we are and the build sequence - tagger-flywheel-implementation-plan, data-centric-tagger-refinement: the feature arc this checkpoint implements Tests: 14 passed (tagger-runs + tag-suggestions).
Builds the human half of the dataset-refinement loop on top of the TagSuggestion model: review near-neighbour disagreements one tag at a time and write fixes back to the labels. Backend (services/tag_suggestion_service.py, routes/tag_suggestions.py): - list/summary/accept/dismiss/reopen plus fix-twin and swap, all undoable (reopen reverses the exact label change: ACCEPTED/TWIN_FIXED/SWAPPED). - Surfaces each pair's picture format (for full-res rendering) and the tagger's per-image confidence. - bulk-accept: decide each pair from the two images' tagger confidence (the near-neighbour link only selects the pair), place it in one of four corners (both / left-only / right-only / neither) and auto-apply the confident ones above a threshold; dry-run returns a sample of the least-confident would-be resolutions to preview, and bulk-reopen batch-undoes. Merges (db_models/tag.py): DEFAULT_TAG_MERGES mirrors pixltagger's tag_remap as an editable constant; the scanner treats a child tag (e.g. "extra digit") as the parent for voting + "missing" eligibility but uses the literal tag for "remove" eligibility. Scanner also gains --tag-remap and --replace. Frontend (ReviewFixesOverlay.vue + useReviewFixesStore.js): a Toolbar- launched overlay framed as "the left is flagged - but is it?", with the 2x2 which-has-it answer (incl. swap), per-image tagger confidence, click- to-zoom/pan, keyboard review, undo, and the bulk-resolve bar with a preview modal. vite.config.js: restore outDir to ../pixlstash/frontend/dist (the in-repo package's served dist) with a comment to prevent the doubled-path mistake.
Tagger-flywheel progress checkpoint, framed by the new design notes. PixlStash as system of record (stats): - TaggerRun model + migration 0059 (chained after 0058_split_caption_sidecars) - POST/GET /tagger-runs: upsert by run name, stores every run incl. rejected, keeps the full report.json payload for the trend/stats panel - Registered in server.py; exported from db_models; 2 API tests Review-queue refinements (on top of 23bdf73): - 4-corner per-image decision in the overlay/store (drop the old direction filter) - pair dedupe in the shared near-neighbour kernel; in-app tag scan; autocomplete Design notes (docs/reviews/): - pixlstash-model-platform-plan: reframes PixlStash as a diffusion dataset & training platform (tagger + LoRA share a GPU-exclusive TrainingJob run via the task manager); records where we are and the build sequence - tagger-flywheel-implementation-plan, data-centric-tagger-refinement: the feature arc this checkpoint implements Tests: 14 passed (tagger-runs + tag-suggestions).
…sh into automation-workflow
* Need this to be able to check whether an golden review set is properly evaluated. * Helps check confidence of tagging decisions and train models
Remove the pin toggle buttons from the sidebar and add a Sidebar Visibility switch in Settings. Persist the preference as sidebar_pinned on the User model (migration 0064) and sync it through the config API.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.