Larq-zoo

:warning: Breaking Changes :warning:

* Drop support for TF 1.14. (156) AdamHillier
* Split zoo into `literature` and `sota` submodules (108) koenhelwegen

:tada: Features

* MeliusNet22 implementation (142) koenhelwegen
* Real-to-Binary implementation (144) timdebruin
* QuickNet implementation (107) jamescook106
* Use AveragePooling2D instead of GlobalPooling2D (128) koenhelwegen

:bug: Bug Fixes

* Add test to ensure that notop-weight are correct (158) lgeiger
* Remove input resolution check for Cifar10 training (129) MariaHeuss
* Move relus outside top (103) koenhelwegen

:book: Documentation

* Add meta info about Larq Zoo to README (122) leonoverweel
* Input shape docstring (117) koenhelwegen
* Fix argument ordering (116) koenhelwegen
* Point to `larq/docs` (113) koenhelwegen
* Fix docstring rendering on the website (114) lgeiger

:construction_worker_man: Internal Improvements

* Add model metrics to docstrings (145) michaeltinsley
* Explicitly set compute softmax in float32 (135) lgeiger
* Sync CI workflows with larq (131) lgeiger
* Restructure non-model files into training and core directories (127) timdebruin
* Remove deprecated layer metrics kwarg (121) lgeiger
* Fix GitHub Actions workflows (104) AdamHillier

:arrow_up: Dependencies

* Larq v0.9 (130) dependabot-preview
* Zookeeper v1.0 (92) AdamHillier

xnornet-v0.2.1
- Fixes bug in `notop` model weights. See 158 for context.

quicknet_xl-v0.1.1
- Fixes bug in `notop` model weights. See 158 for context.
- Removes unnecessary optimizer variables from saved model.

quicknet_large-v0.2.1
- Fixes bug in `notop` model weights. See 158 for context.
- Removes unnecessary optimizer variables from saved model.

quicknet-v0.2.1
- Fixes bug in `notop` model weights. See 158 for context.
- Removes unnecessary optimizer variables from saved model.

meliusnet22-v0.1.1
Fixes bug in `notop` model weights. See 158 for context.

dorefanet-v0.1.1
- Fixes bug in `notop` model weights. See 158 for context.

binary_densenet-v0.1.1
- Fixes bug in `notop` model weights. See 158 for context.

binary_alexnet-v0.2.1
- Fixes bug in `notop` model weights. See 158 for context.

meliusnet22-v0.1.0
Reproduction of MeliusNet22 from [MeliusNet: Can Binary Neural Networks Achieve MobileNet-level Accuracy?](https://arxiv.org/abs/2001.05936).

r2b-v0.1.0

:tada: Features

* Implement QuickNet-XL (137) koenhelwegen
* Implement one-padding and reduce number of SE blocks for QuickNet (136) koenhelwegen

:construction_worker_man: Internal Improvements

* Refactor quicknets and remove duplication (138) koenhelwegen

quicknet_xl-v0.1.0
Introducing QuickNet XL: more neural network for more accuracy.

quicknet_large-v0.2.0
This new version of QuickNet Large includes two changes that improve inference time while maintaining accuracy:
- use of one-padding instead of zero-padding,
- removal of squeeze-and-excite blocks in the first half of the network.

quicknet-v0.2.0
This new version of QuickNet uses one-padding instead of zero-padding, resulting in a faster inference time without any loss in accuracy.

:tada: Features

* Use AveragePooling2D instead of GlobalPooling2D (128) koenhelwegen

:bug: Bug Fixes

* Make min_size in validate_input an argument for Cifar training (129) MariaHeuss
* Change QuickNetLarge model name to "quicknet_large" (124) jneeven

:book: Documentation

* Add meta info about Larq Zoo to README (122) leonoverweel

:construction_worker_man: Internal Improvements

* Sync CI workflows with larq (131) lgeiger
* Restructure non-model files into training and core directories (127) timdebruin
* Remove deprecated layer metrics kwarg (121) lgeiger

:arrow_up: Dependencies

* Update larq requirement from ~=0.8.4 to >=0.8.4,<0.10.0 (130) dependabot-preview

:tada: Features

* QuickNet-Large (112) jamescook106

:bug: Bug Fixes

* Move relus outside top (103) koenhelwegen

:book: Documentation

* Input shape docstring (117) koenhelwegen
* fix argument ordering (116) koenhelwegen
* Point to `larq/docs` (113) koenhelwegen
* Fix docstring rendering on the website (114) lgeiger

:construction_worker_man: Internal Improvements

* Fix GitHub Actions workflows (104) AdamHillier

:arrow_up: Dependencies

* Zookeeper v1.0 (92) AdamHillier

quicknet_large-v0.1.0
This pre-release makes the quicknet-large h5 available for the LCE release.

quicknet-v0.1.0
This pre-release makes the quicknet h5 available for the LCE release.

:warning: Breaking Changes :warning:

* :arrow_up: `larq0.8.0` which drops support for TensorFlow 1.13 (87) lgeiger

:tada: Features

* Add names to all models (88) koenhelwegen

:book: Documentation

* Fix markdown syntax in DoReFa docstring (76) lgeiger

:construction_worker_man: Internal Improvements

* Run unit tests in parallel using `pytest-xdist` (83) lgeiger
* Test against TensorFlow 1.15.0 stable (81) lgeiger
* Use GitHub actions to run unittests and auto-release (80) lgeiger

:arrow_up: Dependencies

* Bump larq to 0.8.1 (90) leonoverweel
* Bump scipy from 1.3.2 to 1.3.3 (89) dependabot-preview
* Bump scipy from 1.3.1 to 1.3.2 (85) dependabot-preview
* Upgrade black (84) lgeiger
* Bump pillow from 6.2.0 to 6.2.1 (82) dependabot-preview
* Bump pillow from 6.1.0 to 6.2.0 (78) dependabot-preview
* Bump tensorflow==2.0.0 (77) lgeiger