GitLab

PiperOrigin-RevId: 370575507

CenterNetMetaArch. It is to make sure the model can be post-quantizable and
doesn't generate calibration errors during quantization.

PiperOrigin-RevId: 370528566

PiperOrigin-RevId: 370255238

* Merged commit includes the following changes:
253126424  by Andre Araujo:

    Scripts to compute metrics for Google Landmarks dataset.

    Also, a small fix to metric in retrieval case: avoids duplicate predicted images.

--
253118971  by Andre Araujo:

    Metrics for Google Landmarks dataset.

--
253106953  by Andre Araujo:

    Library to read files from Google Landmarks challenges.

--
250700636  by Andre Araujo:

    Handle case of aggregation extraction with empty set of input features.

--
250516819  by Andre Araujo:

    Add minimum size for DELF extractor.

--
250435822  by Andre Araujo:

    Add max_image_size/min_image_size for open-source DELF proto / module.

--
250414606  by Andre Araujo:

    Refactor extract_aggregation to allow reuse with different datasets.

--
250356863  by Andre Araujo:

    Remove unnecessary cmd_args variable from boxes_and_features_extraction.

--
249783379  by Andre Araujo:

    ...

logic doesn't depend on the per_keypoint_offset value and use
get_batch_predictions_from_indices to be consistent with other prediction
heads.

PiperOrigin-RevId: 369925004

PiperOrigin-RevId: 369749466

* OD tutorial colab working

* fixing install instructions link

Surgical update to CenterNet keypoint postprocessing to use bounding boxes if present. This is shown to reduce keypoint snapping across individuals.

PiperOrigin-RevId: 369432018

* Dataset utilities added.

* Global model definition

* Dataset modules added.

* Dataset modules fix.

* global features model training added

* global features fix

* Test dataset update

* PR fixes

* repo sync

* repo sync

* Syncing 2

* Syncing 2

* Added global model supporting modules

* code style fixes

* Minor style fixes

Call set_keypoint_visibilities on the non-expanded versions of the detection and groundtruth keypoints. set_keypoint_visibilities expects a rank-3 tensor, and was being provided a rank-4 tensor. This had the unintended effect of creating a keypoint visibilities tensor of the wrong shape, resulting in only 2 keypoints being visualized.

PiperOrigin-RevId: 368458273

Fix the typo-ed name "remove_unecessary_ema" -> "remove_unnecessary_ema" and also re-implement it with deep dictionary copy.

PiperOrigin-RevId: 368338501

* Dataset utilities added.

* Global model definition

* Dataset modules added.

* Dataset modules fix.

* global features model training added

* global features fix

* Test dataset update

* PR fixes

* repo sync

* repo sync

* modules moving

* Removed unnecessary modules

* Removed unnecessary files

PiperOrigin-RevId: 368217758

PiperOrigin-RevId: 368215027

This replaces symbols in tf.keras.mixed_precision.experimental with the corresponding nonexperimental symbols. In some cases, passing a Policy is replaced with passing a policy name for conciseness.

Additionally, for the Shakespeare model, the loss_scale flag is removed, since supporting it with the nonexperimental API is slightly more verbose and it is recommended users use the default loss scale.

PiperOrigin-RevId: 368123944

PiperOrigin-RevId: 368024164

Rewrite the GIOU loss op in matrix form to avoid the map_fn. map_fn could be prohibitively costly when dealing with large number of bounding boxes.

PiperOrigin-RevId: 366854232

PiperOrigin-RevId: 366850188

PiperOrigin-RevId: 366395416

PiperOrigin-RevId: 366388829

PiperOrigin-RevId: 366385685

PiperOrigin-RevId: 366382220

PiperOrigin-RevId: 366134194

PiperOrigin-RevId: 365913722

PiperOrigin-RevId: 365899795

This CL register a pre-defined GIOU loss to the losses.proto and losses_builder so that the GIOU loss can be correctly called & built when GIOU loss is specified in the loss config file.

PiperOrigin-RevId: 365891022

PiperOrigin-RevId: 365683210

PiperOrigin-RevId: 365177695

PiperOrigin-RevId: 364924351

PiperOrigin-RevId: 364895608

PiperOrigin-RevId: 364731599

PiperOrigin-RevId: 364671967

Updated the single instance postprocessing logics such that it uses the score to distance ratio scoring method. The new logics also avoid using expensive ops, e.g. reduce_max/maximum.

PiperOrigin-RevId: 364607976

PiperOrigin-RevId: 364388006

PiperOrigin-RevId: 364363163

PiperOrigin-RevId: 364349645

PiperOrigin-RevId: 363975731