GitLab

PiperOrigin-RevId: 319058156

318938278  by Andre Araujo:

    Loading pretrained ImageNet weights to initialize the ResNet backbone. Changed the defaults of the batch size and initial learning rate to increase convergence on the GLDv2 dataset. Made the evaluation batch size dynamic depending on the global batch size.

--
318911740  by Andre Araujo:

    Introduced additional shuffling of the TRAIN and VALIDATION datasets to ensure label variance across batches.

--
318908335  by Andre Araujo:

    Export model migration to TF2.

--
318489123  by Andre Araujo:

    Model exporting script for global feature trained with DELF codebase.
    Additionally, makes a small change to replace back_prop=False in the tf.while_loop call (see deprecation notice in https://www.tensorflow.org/api_docs/python/tf/while_loop

).

--
318401984  by Andre Araujo:

    Add attention visualization to DELF training script.

--
318168500  by Andre Araujo:

    Several small changes to DELF open-source training code:
    - Replace "make_dataset_iterator" call which was deprecated by a more recent suitable version.
    - Add image summary, allowing visualization of the augmented images during training
    - Normalize images before feeding them to the model

--
316888714  by Andre Araujo:

    - Removed unnecessary cast from feature_aggregation_extraction.py
    - Fixed clustering script

--

PiperOrigin-RevId: 318938278
Co-authored-by: Andre Araujo <andrearaujo@google.com>

PiperOrigin-RevId: 318945656

PiperOrigin-RevId: 318942343

PiperOrigin-RevId: 318935091

Contributed by mickeystroller

PiperOrigin-RevId: 318909403

PiperOrigin-RevId: 318851715

PiperOrigin-RevId: 318829996

PiperOrigin-RevId: 318755856

PiperOrigin-RevId: 318714418

318569851  by jonathanhuang:

    Fix for fine-tuning from classification checkpoints in the v2 binary.

--

PiperOrigin-RevId: 318569851

318545448  by jonathanhuang:

    Modifies visualization code in TF2 evaluation loop so that we don't write out image summaries to disk for every single image.  This change will reduce summary file sizes by ~2 orders of magnitude on average and speed up evaluation cycles (20 minutes per COCO eval cycle vs 2 hours for RetinaNet).

--
318514741  by sbeery:

    Adding link to the blog post

--

PiperOrigin-RevId: 318545448

318497061  by rathodv:

    1. Replace strategy.run() with strategy.experimental_run_v2() and replace tensor.ref() with tensor.experimental_ref() to be compatible with TF2.1 runtime on cloud.
    2. Fix expected string in failing PY3 tests.

--
318493408  by aom:

    Implements "Bidirectional Feature Pyramid Network Generators" for BiFPN-based feature extractors (e.g. EfficientDet).

--

PiperOrigin-RevId: 318497061

318417714  by jonathanhuang:

    Internal change.

--
318367213  by sbeery:

    Pointing users to more documentation for beam

--
318358685  by sbeery:

    Context R-CNN sample config for GPU

--
318309800  by rathodv:

    Internal

--
318303364  by ronnyvotel:

    Adding the option for parsing and including DensePose annotations. http://densepose.org/

--
318291319  by aom:

    Adds conv_bn_act conv_block option, and naming convention changes for BiFPN utils.

--
318200598  by ronnyvotel:

    Updating the TF Example Decoder to parse DensePose annotations.

--
318174065  by jonathanhuang:

    Internal change.

--
318167805  by rathodv:

    Add use_tpu flag to TF2 binary.

--
318145285  by aom:

    Adds option for convolutional keras box predictor to force use_bias.

--

PiperOrigin-RevId: 318417714

PiperOrigin-RevId: 318421049

PiperOrigin-RevId: 318387106

PiperOrigin-RevId: 318322629

PiperOrigin-RevId: 318208409