AI Showcases - Center for Aritifical Intellgience and Robotics (CAIRO)
This repository contains the code for the AI use cases of the Center of Artificial Intelligence and Robotics (CAIRO)
- Jukebox (Music Embedding, 492 songs 15 artists)
- Twitter (Twitter Dataset Embedding ~60k Tweets from biggest Nasdaq companies.)
- Style Transfer (Transfer style to arbitrary image.)
├── Jukebox # Mozart Jukebox app to visualize music embeddings. See:
│ ├── docs # Explanations for jukebox
│ ├── index.html # Main Webapp File
│ ├── data_storage # Storage of trained embedding for mozart
│ ├── deploy.sh # deployment file on the showcase server
│ └── others # Favicons, licences etc.
├── style # Style Transfer application. See:
│ ├── docs # Explanations for jukebox
│ ├── index.html # Main Webapp File
│ ├── data_storage # Storage of trained embedding for mozart
│ ├── others # Development files, favicons etcs.
│ ├── deploy.sh # deployment file on the showcase server
│ └── saved_model_* # Saved Tensorflow.Js models
├── Twitter # Twitter Embedding to visualize our Twitter dataset embeddings. See:
│ ├── index.html # Main Webapp File
│ ├── data_storage # Storage of trained embedding for mozart
│ ├── deploy.sh # deployment file on the showcase server
└── └── others # Favicons, licences etc.
The Embeddings for Mozart Jukebox and Twitter are stored under <app>/data_storage. New embeddings must also be stored there. Finally, the new embedding must be registered in <app>/data_storage/projector_config.json. How is done is self-explaining in the file itself.
To save data comaptible with the tensorboard used at Jukebox or Twitter your data has to be float32 an in *.bytes format.
This can achieved by:
# Save data
features.dtype=np.float32
features.tofile(board_data+data_path)All applications can be hosted via a simple webserver. In the deploy.sh is shown how to deploy to apache webserver.
The Twitter webapp is a yarn application. To build the app use:
yarn run prep
yarn run build
The build applications can be inspected for development via:
yarn run start
- Style app by https://github.com/reiinakano
- Twitter Embedding trained by Julian Hanke, Christian Niekler, Trong Duc Tran
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent