94

GitHub - EliotAndres/pretrained.ml: Compilation of pre-trained deep learning mod...

 6 years ago
source link: https://github.com/EliotAndres/pretrained.ml
Go to the source link to view the article. You can view the picture content, updated content and better typesetting reading experience. If the link is broken, please click the button below to view the snapshot at that time.

[DEPRECATED] pretrained.ml

[DEPRECATED] Sortable and searchable compilation of pre-trained deep learning models. With demos and code.

DEPRECATED: You can find an alternative on modeldepot.io (with live demos) or on modelzoo.co

A word of warning

This is running on a server without GPU, hence it seems slow.

Also, the code may look a bit like monkey-patching for the following reasons:

  • Models are cloned as submodules: therefore we have to mess around with the python path :-(
  • There is a queuing systems for the jobs (allows user to see their job's position in the queue)

About

Having spent too much time installing deep learning models just to evaluate their performance, I created this repo for several reasons:

  • Access a free demo of deep learning models
  • Gather available deep learning models
  • Get a docker container running the model for a quick install

Installation

Requirements: Docker, docker-compose and enough space free for the model weights.

git clone https://github.com/EliotAndres/pretrained.ml --recursive
cd containers
docker-compose build
docker-compose up -d

Useful commands

docker ps #list images
docker attach [container_id] #attach a shell to specific image

Contributing

Many models are missing. Any help is welcome ! You have two options to contribute.

Easy way: Add a model to the list without a demo:

  • Fork the repo
  • Edit the docs/models.yaml file (you can even edit it with Github's editor)
  • Make a pull request

Other way: add a model with a demo:

  • Fork the repo
  • Add the model inside one of the docker containers
  • Create a route in the serve.py file
  • Add a demo calling the route
  • Make a pull request

Todos

  • Use nvidia-docker ?
  • Add flag to compile Tensorflow
  • Consider splitting each model in a different container ?
  • Linter
  • Add analytics

About Joyk


Aggregate valuable and interesting links.
Joyk means Joy of geeK