GitLab

--
d9b85094711b4bcd34400989618b738032c992cf by Srihari Humbarwadi <sriharihumbarwadi97@gmail.com>:

Added dataloader for Panoptic Mask R-CNN (#10112)

* Added `panoptic_maskrcnn_input.Parser`

* Added tests for `panoptic_maskrcnn_input.Parser`

* added `DataConfig`

* fixed doc-string

* Added type annotations

COPYBARA_INTEGRATE_REVIEW=https://github.com/tensorflow/models/pull/10144 from tensorflow:panoptic-segmentation d9b85094711b4bcd34400989618b738032c992cf
PiperOrigin-RevId: 386126918

PiperOrigin-RevId: 386105413

PiperOrigin-RevId: 386029066

PiperOrigin-RevId: 385674527

PiperOrigin-RevId: 385672449

PiperOrigin-RevId: 385652538

PiperOrigin-RevId: 385629372

PiperOrigin-RevId: 385615968

PiperOrigin-RevId: 385611054

PiperOrigin-RevId: 385593727

Both unit tests and real use cases are passing None to it.

PiperOrigin-RevId: 385468558

PiperOrigin-RevId: 385256430

PiperOrigin-RevId: 384957416

[ranking] Passing `experimental_prefetch_to_device=False` option for distributed dataset of Ranking models.

PiperOrigin-RevId: 384792951

PiperOrigin-RevId: 384509754

PiperOrigin-RevId: 384478988

PiperOrigin-RevId: 384358378

PiperOrigin-RevId: 384356677

PiperOrigin-RevId: 384262412

Fix the issue when both numerator and denominator are 0, the dice score is incorrectly computed as 1, which should be 0.

PiperOrigin-RevId: 384257731

PiperOrigin-RevId: 384257276

PiperOrigin-RevId: 384018258

[efficient] Fix is_short_seq order so that we can also apply feature transform and apply softmax afterwards.

PiperOrigin-RevId: 383967806

PiperOrigin-RevId: 383944185

PiperOrigin-RevId: 383861528

PiperOrigin-RevId: 383856955

Before, it was too easy to accidentally forget to set runtime.loss_scale, which had to always be done if mixed precision is used, otherwise the model would converge to worse accuracy. Now, all that needs to be done to use mixed precision is to set runtime.mixed_precision_dtype=float16.

PiperOrigin-RevId: 383767033

PiperOrigin-RevId: 383679175

PiperOrigin-RevId: 383667471

PiperOrigin-RevId: 383660517

PiperOrigin-RevId: 383505893

PiperOrigin-RevId: 383505455

PiperOrigin-RevId: 383066915

PiperOrigin-RevId: 382960239

PiperOrigin-RevId: 382846192

PiperOrigin-RevId: 382833629

PiperOrigin-RevId: 382655843

Added use_keras_upsampling_2d flags to use keras upsampling layer instead of optimized custom spatial_transform_ops.

PiperOrigin-RevId: 382626602

PiperOrigin-RevId: 382606601

PiperOrigin-RevId: 382571340