feat: annotation platform dataset and multi-dataset training#17
Merged
ziv-lazarov-nagish merged 8 commits intonagishfrom Apr 14, 2026
Merged
feat: annotation platform dataset and multi-dataset training#17ziv-lazarov-nagish merged 8 commits intonagishfrom
ziv-lazarov-nagish merged 8 commits intonagishfrom
Conversation
AmitMY
reviewed
Apr 14, 2026
Contributor
AmitMY
left a comment
AmitMY
left a comment
There was a problem hiding this comment.
multi dataset support should be available in main - maybe we pass datasets = 'dgs,platform' etc in args -
then ideally, we should have a register_dataset("dgs", some_function) and we can register datasets - then not necessary to have specific code for each dataset everywhere
Contributor
Author
i'll implement the comma-separated datasets names but i'm not sure i understand the |
3 tasks
Add annotation platform dataset backed by Convex DB sync: - sync.py: fetch completed annotations (completedAt + approved/submitted) - dataset.py: hash-based train/dev/test splits, quality filtering - Multi-dataset dispatch (--dataset platform/combined) - --val_dgs flag to mix DGS dev into validation - --annotations_path, --quality_percentile args - [platform] optional dep group (httpx, python-dotenv) - Split manifest tracking for reproducibility
Replace broken --score boolean flag (defaulted to True, making sync unreachable) with proper subcommands. Update run_hpo.sh to use the CLI directly instead of python -c workaround.
Add --dataset flag (dgs/platform) and --annotations_path arg so the evaluation script can run on both DGS and annotation platform data.
Remove approved status filter — all production workflows have no review stage, so every completed status is treated as good.
Remove subcommands — single CLI that syncs then scores by default. Add --no_score to skip scoring. Skip sync if cache file exists, skip already-scored videos. Fix gpu->cuda device mapping for checkpoint loading.
ziv-lazarov-nagish
force-pushed
the
feat/annotation-platform-dataset
branch
from
967b76a to
9e344cd
Compare
AmitMY
approved these changes
Apr 14, 2026
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
2 participants
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.
Summary
platformvia dataset registry (--datasets platformor--datasets dgs,platform)--quality_percentilefiltering for annotation quality--annotations_patharg for cache locationcompletedAt), scores with latest model, caches resultsChanged files
sign_language_segmentation/datasets/annotation_platform/— dataset, sync, and testssign_language_segmentation/args.py—--annotations_path,--quality_percentilesign_language_segmentation/datasets/common.py— register platform in_ensure_datasets_registeredsign_language_segmentation/evaluate.py—--annotations_pathargpyproject.toml—[platform]optional group (httpx, python-dotenv)Test plan
ruff check .passespytestpasses (61 tests)--datasets platformfine-tune (hm_iou: 0.469)--datasets dgs,platformfrom scratch (hm_iou: 0.391)--datasets dgs,platformfine-tune (hm_iou: 0.457)