Proposing CVPM, the Computer Vision Package Manager


(Xiaozhe Yao) #1

Dear OSC Teams,

Hi, I am the maintainer of CVPM, which is the abbreviation of Computer Vision Package Manager. It is designed to help developers to easily deploy machine learning models (currently computer vision models) with simple commands.

The main use cases include:

  • For researchers, they could use CVPM to verify others’ result. When they are proposing an academic papers, a model file can be attached to the paper so that others can verify it with our web interfaces easily. It is different from other tools like pip because it hides many implementation details and can be used more smoothly.

  • For developers, CVPM could help them to build the HTTP interface between clients and their algorithm. Many algorithm and deep learning developers are not familiar with web programming and this tool can therefore benefit them. With CVPM, they can focus on writing their algorithms and do not need to worry about the deployments of models.

  • We are also building a hub for model sharing and trade. We hope it will be able to help indie developers to sell their deep learning/machine learning models online to commercials. And we think it will become one of our profitable point.

Our project is released under MIT License, which is quite loose we think. The homepage is at https://github.com/unarxiv/cvpm. It is still at startup phase and under heavy development. Our project (includes logos, and etc) have no patents held by others.

We are looking forward to joining the community and working together with the community. And therefore we propose this project to become a member of OSC Project.

Thanks for you reading. If there’s any concern and questions, please reply below and I will answer it as soon as possible.


(Xiaozhe Yao) #2

any comments or ideas? :smiley: I have sent a simple email but not received any reply yet…


(Rob Worthington) #3

Hi Xiaozhe,

I like your idea for sharing models. That will be extremely useful as this field develops. I hope you are able to develop that further as a way of funding your other work.

Which do you think are the most interesting models available on your hub?

Rob


(Xiaozhe Yao) #4

Hi Rob,

Thanks for your comment and support! We have made great progress since the last post, we hope we will publish a next release (include a very easy-to-use web ui) to help our users get started. By then I will add it to this post!

In all models, which vary from face detection to NSFW image recognition, I think the NSFW recognition is the most interesting one for me :smiley:

Hope you and your projects are working well!

Xiaozhe