Paddlehub

[1、Support text-to-image domain model]
- Add five text-to-image domain models based on disco diffusion, in which three models are for English and two for Chinese. Especially, Chinese text-to-image model [disco_diffusion_ernievil_base](https://aistudio.baidu.com/aistudio/projectdetail/4444998) is based on Baidu **ERNIE-ViL**，welcome to experience.

【2、Support Wenxin large models API】
- Add api call for [**ERNIE-ViLG**](https://aistudio.baidu.com/aistudio/projectdetail/4445016) model, which supports text-to-image task。
- Add api call for [**ERNIE 3.0 Zeus**](https://aistudio.baidu.com/aistudio/projectdetail/4445054) model, which supports applications such as writing essays, summarization, couplets, question answering, writing novels and completing text.

[ 1. Improvements]

- Add supports for five new models, including two high-precision semantic segmentation models based on VOC dataset and three voice classification models.
- Enforce the transfer learning capabilities for image semantic segmentation, text semantic matching and voice classification on related datasets.

[ 2. Upgrades of deployment capabilities]

- Add the export function APIs for two kinds of model formats, i.,e, ONNX and PaddleInference.
- **Important Open-Source Ecological Cooperation**: add the support for [BentoML](https://github.com/bentoml/BentoML/), which is a cloud native framework for serving deployment. Users can easily serve pre-trained models from [PaddleHub](https://github.com/PaddlePaddle/PaddleHub) by following the [Tutorial notebooks](https:// github.com/PaddlePaddle/PaddleHub/tree/release/v2.1/demo/serving/BentoML). Also, see this announcement and [Release note](https://github.com/bentoml/BentoML/releases/tag/v0.12.1) from BentoML. (Many thanks to [parano](https://github.com/parano) [cqvu](https://github.com/cqvu) [deehrlic](https://github.com/deehrlic) for contributing this feature in PaddleHub)

[ 3. Bug fixes ]

- [7da1230](https://github.com/PaddlePaddle/PaddleHub/commit/7da12302dd77e3d739da72821d41715ad8a7c79c) Fixed the problem that the model cannot resume training if metrics is not recorded.
- [b0b3144](https://github.com/PaddlePaddle/PaddleHub/commit/b0b3144eff34e47cac8fc450c8b7cb6c557f9b84) Fixed the problem that the thread did not exit normally when the evaluation process was abnormal.
- [30aace4](https://github.com/PaddlePaddle/PaddleHub/commit/30aace46414bbeef02beb75b7128f48fada82150) Improve the model installation process.

* 发布 2.0版本，全面迁移动态图编程模式，模型开发调试更加方便，finetune接口更加灵活易用。
* 视觉类任务迁移学习能力全面升级，支持图像分类、图像着色、风格迁移等多种任务。
* BERT、ERNIE、RoBERTa等Transformer类模型升级至动态图，支持文本分类、序列标注的Fine-Tune能力。
* 新增词向量模型61个，其中包含中文模型51个，英文模型10个。
* 优化服务化部署Serving能力，支持多卡预测、自动负载均衡，性能大幅度提升。
* 新增自动数据增强能力Auto Augment，能高效地搜索适合数据集的数据增强策略组合。

* BERT、ERNIE、RoBERTa等Transformer类模型升级至动态图，增加[文本分类](../../demo/text_classification)的Fine-Tune能力
* 修复部分已知问题

* 全面迁移动态图编程模式，模型开发调试更加方便，finetune接口更加灵活易用。
* 优化服务化部署Serving能力，支持多卡预测、自动负载均衡，性能大幅度提升。
* 视觉类迁移学习能力全面升级，支持[图像分类](../../demo/image_classification)、[图像着色](../../demo/colorization)、[风格迁移](../../demo/style_transfer)等多种视觉任务。

* 『[图像分割](https://www.paddlepaddle.org.cn/hublist?filter=en_category&value=ImageSegmentation)』新增轻量级人像分割模型Humanseg，支持移动端实时分割
* 增强文本匹配任务性能，使用[EMNLP2019-Sentence-BERT](https://arxiv.org/abs/1908.10084)作为文本匹配任务网络，可同时大幅提升准确率和预测速度。配套教程：[pointwise文本语义匹配](https://aistudio.baidu.com/aistudio/projectdetail/705526)、[pairwise文本语义匹配](https://aistudio.baidu.com/aistudio/projectdetail/709472)
* 修复文本分类选择F1作为评价指标，运行错误