Fastestimator

Dataset
* ExtendDataset: extending dataset length for better performance

Pipeline
* BatchOp: more flexible batching behavior and support batched `NumpyOp`
* FilterOp: removing sample elements based on criteria

Network
* l2 regularization as TensorOp: allow for tf model to use equivalent weight decay as torch

Estimator
* new `train_steps_per_epoch` and `eval_steps_per_epoch` argument. Now user can truly set arbitrary epoch length with them.

Apphub
* Instance segmentation - SOLOv2 TensorFlow & PyTorch implementation

Tutorial
* Custom Data Loaders
* Finetuning

Others
This release also includes many bug fixes throughout the API.

This release is a patch of FE1.3.X releases about sklearn and pycocotool dependency.

* Fixed a critical bug about database access

* Fixed a critical bug when setting `num_process=0` in pipeline and printing together

* New History API
* New Pipeline num_process defaults

New System
* Multi dataset support throughout the API. [example](https://github.com/fastestimator/fastestimator/blob/master/tutorial/advanced/t13_multi-dataset_training_evaluation.ipynb)

Dataset
* SVHN Cropped classification dataset
* Enable nested list in dataset and BatchDataset. [example](https://github.com/fastestimator-util/fastestimator-misc/tree/master/examples/1.3/special_batch_compo)
* TEDLR NMT dataset

Pipeline
* Enhance Pipeline key filtering warning
* Change pipeline.get_loader to context manager (non backward-compatible). [example](https://github.com/fastestimator-util/fastestimator-misc/blob/master/examples/1.3/pipeline_get_loader/get_loader.py)
* Integrated new augmentation methods: `RandomShape`, `IAAAdditiveGaussianNoise`, `IAACropAndPad`

Network:
* Auto Weight decay adjustment for SGDW related optimizers in tensorflow addons
* CrossEntropy loss can apply class weights now. [example](https://github.com/fastestimator-util/fastestimator-misc/tree/master/examples/1.3/cross_entropy_class_weight)
* ModelOp can extract intermediate layer results easily. [example](https://github.com/fastestimator-util/fastestimator-misc/tree/master/examples/1.3/cross_entropy_class_weight)
* New attentionUnet architecture is now available for direct import.
* Start using experimental_relax_shapes=True in to make different batch shape training more efficient for tensorflow. (now tf only builds graph when rank of input changes, as opposed to when shape changes)
* TensorFlow user can have the option to save architecture together with weights.
* Added fine-tuning example. [example](https://github.com/fastestimator-util/fastestimator-misc/tree/master/examples/1.3/pretrain_finetune)

Estimator

* Enable eager option in tensorflow backend for easier debugging [example](https://github.com/fastestimator-util/fastestimator-misc/tree/master/examples/1.3/eager_warmup)

XAI:
* UMAP
* Eigen Cam. [example](https://github.com/fastestimator/fastestimator/tree/master/tutorial/xai)
* Grad Cam. [example](https://github.com/fastestimator/fastestimator/blob/master/tutorial/xai/t05_gradcam.ipynb)

Apphub
* Super Convergence
* NASWOT
* YOLOV5
* Transformer
* Vision Transformer