Faster/Mask R-CNNN works on mobile (Android)?

[hachreak]

❓ Questions and Help

Hi everybody,
I was trying understand if I can make object detection on Android but it's not really clear if I can do it.
The most recent comments that I found are these ones:

• https://discuss.pytorch.org/t/support-torchvision-ops-nms-on-mobile-cpu/69627/2

• Attempted to read a PyTorch file with version 3, but the maximum supported version for reading is 2. Your PyTorch installation may be too old. (init at ../caffe2/serialize/inline_container.cc:132) #1943 (comment)

Can somebody confirm it?
If not, maybe we can leave the issue open until vision is ready for mobile?
Thanks 😄

[jjbouza]

I'm also interested in this

[fmassa]

It is not yet supported, but it is in the plans to be supported in follow-up releases of PyTorch

[lothlorien13]

Is it currently supported?

[ValerianGonnot]

@lothlorien13 No, it is still not supported in Pytorch 1.7.0

[fmassa]

@ValerianGonnot note that once we merge (which should be soon) #2897 torchvision detection models will be supporting Android

[fmassa]

The next release of torchvision (which will be released in ~1 month) will have support for Android for faster rcnn / mask rcnn.

Here is a demo app to show how to do it https://github.com/pytorch/vision/tree/master/android/test_app