Detectron2: FBnet backbone

[startupgurukul]

fbnet backbone for 🚀 like inference

Edge AI is always awesome and unlocks solutions for lot of humanitys problems

Motivation

Considering mobile processors are offering so much power via their additional NN processing Unit, detectron2 mobile friendly models can empower building apps which can run locally and can empower internet limited users aswell

Pitch

this has been on the roadmap, this issue is a placeholder

[startupgurukul]

Fbnet + Pointrend accuracy improvements would be interesting to observe.

[youngwanLEE]

@startupgurukul

I implemented efficient backbone networks, _VoVNet_, for detectron2.

[Samjith888]

@startupgurukul

I implemented efficient backbone networks, _VoVNet_, for detectron2.

Please maintain repo and resolve the issues which is open in ur repo

[breznak]

@youngwanLEE would you try to merge your models to Detectron2 here?

[youngwanLEE]

@breznak

We already tried to merge our vovnet like this.

[Dorozhko-Anton]

@startupgurukul @ppwwyyxx Would you like an implementation from scratch and then you will train it or an implementation compatible with maskrcnn_benchmark weights for fbnet ?

[startupgurukul]

Any love for this folks, supposed to ETA was around 1-2 weeks when issue was created.

https://github.com/facebookresearch/detectron2/issues/12#issuecomment-565566046