Add vision models

[darsnack]

Create standard models for vision tasks.

Reference issue: darsnack/FluxModels.jl#1

[darsnack]

If we go with #3 (comment) then it seems like many of the other models will reside in separate packages. So maybe we can do the same for the vision models.

[darsnack]

Current progress:

• https://github.com/darsnack/Metalhead.jl/tree/darsnack/vision-refactor
• Refactor repo with more complete models and documentation Metalhead.jl#70

Accepting help to finish this up.

[cyborg1995]

Hello

I would love to work on the squeezenet model.
I'm new to julia and to github but I'm familiar with dl using PyTorch. Could someone guide me on how to start working on this?
Also, which repository to fork?

Thanks

[darsnack]

Hi @cyborg1995. The best way to help with this issue is to push FluxML/Metalhead.jl#70 across the finish line. Most of the models are complete, but we need to train them and add the BSON weights. I will post more tomorrow on how to add the weights to the repo, but for now, it would be a big help if you can clone the fork and attempt to train each model to full accuracy. I think this would be a great way to help out and get familiar with Flux+Julia.

[darsnack]

Closing this since Metalhead has been refactored to address this concern.