For setting up Azure, I've uploaded all the required files to Azure Machine Learning Datastores. Whenever the data is required, it is referred from the respective Datastore.
If you would like to set things up locally, the following links are useful
Haikus for Language Model
- Cleaned haiku dataset: https://bit.ly/2MHlFMu
AttnGAN
- Image Encoder - https://bit.ly/2MGBnaH
- Text Encoder - https://bit.ly/2ZA72zX
- AttnGAN Model - https://bit.ly/2F34vEZ
- COCO dataset - https://bit.ly/367Ttdn
I've written a convenient clone of AttnGAN here
Refer the repo for setting up all these requirements using a single shell script. This is the easier way, you may try setting things up from the official repo as well.
GPT-2 finetuning
- GPT-2 finetuning - https://github.com/minimaxir/gpt-2-simple
FastPhotoStyle
- FastPhotoStyle - https://github.com/NVIDIA/FastPhotoStyle
AttnGAN
Code of a convenient clone of AttnGAN by MSFT Research. AttnGAN original repo contains the training and testing code. But this folder is solely intended to use for generating images from text using MSCOCO pretrained weights.
- Original paper : http://openaccess.thecvf.com/content_cvpr_2018/papers/Xu_AttnGAN_Fine-Grained_Text_CVPR_2018_paper.pdf
- Original repo : https://github.com/taoxugit/AttnGAN
Dockerfiles
Dockerfiles and requirements required for building various docker containers needed to build Auria's pipeline steps.
GPT_2_Finetuning
Code of convenient clone of finetuning the GPT-2
model using 117M pretrained model and then generate text from it.
- Original paper: https://d4mucfpksywv.cloudfront.net/better-language-models/language-models.pdf
- Original repo: https://github.com/minimaxir/gpt-2-simple
Notebooks
Different Jupyter Notebooks
used to build training, testing and building pipelines of Auria.
NVIDIAFastPhotoStyle
Code of convenient clone of FastPhotoStyle by NVIDIA.
- Original paper: https://arxiv.org/abs/1802.06474
- Original repo: https://github.com/NVIDIA/FastPhotoStyle
-
You'll need an Azure ML Pipelines subscription and its configuration JSON file to use the platform. You can refer the notebooks to know the usage of these config files.
-
Also install the
azure_requirements.txt
.
pip3 install -r azure_requirements.txt