[FEATURE] OpenImages Pretrained Models

[rsomani95]

Hi Ross, thanks for sharing this repo!
Curious if you have any plans on releasing a model trained on OpenImages?

Is your feature request related to a problem? Please describe.
No

Describe the solution you'd like
Have access to a pretrained model trained on OpenImages

Describe alternatives you've considered
The official repo (see here), but their model is outdated and it's also in tensorflow, which is inconvenient

Additional context
N/A

[rwightman]

@rsomani95 that has been the goal, I did train one once (> 1 year ago) on a faster-rcnn type net w/ first version of my dataset code, but not with these nets yet. It takes a bloody long time so I've been waiting until I have more access to more compute. I might be able to do it on TPUs soon, but getting this working on PyTorch XLA could be a bit of work.

[rsomani95]

@rwightman Thanks for the response. Great to know it's somewhere on your roadmap.
I'd love to have helped but have no experience PyTorch XLA and am unlikely to gain access to TPUs anytime soon.

[jdsannchao]

Hi Ross,
Thanks for sharing this repo!
May I ask how is the plan going, were you able to train the model on OpenImages?

[rwightman]

@jdsannchao no not yet, I only have 2x GPU machines free right now and it'd probably take close to two months for a decent sized model. Getting closer to the point where I think PyTorch XLA might work though... but have more to do still