Symmetry-aware flip augmentation #455
Merged
Conversation
This file contains 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
- Make sure indices are all cast to int32
talmo
added a commit
that referenced
this pull request
Mar 24, 2021
* Add track indices to instance cropper * Add class vector generator * Split class vectors correctly in instance cropper * Move head output layer construction to heads module - Heads now subclass a base Head class - Naming doesn't include _0 anymore since we don't have any multi-output models for now. - Better input validation in Model.from_config constructor - Add loss weight to all heads in config - Test coverage for heads and (minimally for) model * Add topdown config, head and model - Rename multiclass to multiclass_bottomup * Add trainer * Data pipeline * Apply black to 'sleap' and 'tests' (#465) Co-authored-by: Arie Matsliah <[email protected]> * Fix model creation and add pooling param to head * Symmetry-aware flip augmentation (#455) * Implement symmetry-aware instance reflection * Fix symmetries sometimes not being returned uniquely * Add fancier indexing to instances * Add random flipping transformer * Fix failing linux test - Make sure indices are all cast to int32 * Add vertical flip * Add flip augmentation to config, GUI and pipeline builders * Update profiles with default fields Co-authored-by: ariematsliah-princeton <[email protected]> * Multi-size videos in data pipelines (#440) Add support for variable size videos within the same dataset by matching their size with padding or resizing Co-authored-by: Arie Matsliah <[email protected]> * Type check + Lint in CI (#470) * Try lint and typecheck in CI workflow * update * nit * continue on MyPy errors * test * correct * correct * correct Co-authored-by: Arie Matsliah <[email protected]> * Rename predictors for consistency with inference layers - TopdownPredictor -> TopDownPredictor - BottomupPredictor -> BottomUpPredictor * Create PULL_REQUEST_TEMPLATE.md * Update authors list (#471) Co-authored-by: Arie Matsliah <[email protected]> * Add CLA (#473) * Add CLA * update links Co-authored-by: Arie Matsliah <[email protected]> * Update PULL_REQUEST_TEMPLATE.md * Miscellaneous QOL (#467) Pre-1.1.0 update features (changelist in #467) * Bump pre-release version * Add back load_model that got lost in the merge - Add detection of bottomup and topdown multi-class model loading * Fix more missing things post-merge * Fix lint * Fix training from config * Add inference * Tweak describe tensor to accept nested tuples/dicts * Lint * Fix test * Lint * Fix load video dataset arg * Fix inference * Fix evals * Add BU MC to evals * Remove batch norm from TD MC head * Add option to disable batch norm in pretrained models * Add track matching when merging labels * Don't error when training finishes with no inference target Co-authored-by: ariematsliah-princeton <[email protected]> Co-authored-by: Arie Matsliah <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
This PR implements horizontal flipping augmentation. This is tricky in our setting since we have to swap left/right symmetric nodes in addition to adjusting the coordinates.
Addresses the long-standing #228.