lupesko / model-zoo-old Goto Github PK

The top level classifications on the landing page mix two different things with results that make it difficult to navigate:

1. research domain or application (e.g. NLP, Image classification)
2. model architecture

I would suggest that the top level organization be research application only. Generative models or autoencoders are a low-level implementation thing -- model architectures you can use to solve various problems so put those models under the problem domain.

Also, traditionally ASR and TTS would not be considered NLP (they have entirely different conferences and journals for example).

A possible high level organization might be:

Computer vision / Computer graphics

• Image detection
• Semantic segmentation
• Face detection
• Super resolution
• Style transfer
  ...

NLP

• NMT
• Language modeling

Speech

• ASR
• TTS (Deep Voice, Wavenet etc)

Vision + NLP

• Visual question answering etc