search

tsubauaaa / flutter_d2go

Flutter plugin that runs d2go
BSD 3-Clause "New" or "Revised" License
10 stars 10 forks source link

Which torchvision versions should I train on for this plugin? #13

Open jimjam-slam opened 1 year ago

jimjam-slam commented 1 year ago

Hi there! very excited to use this plugin to do object detection. The only thing is, I'm not sure which version of torch and torchvision to use to train my models. In android/build.gradle, it looks like you're using Torchvision Lite 1.11, while in the ios/flutter_d2go.podspec, it looks like it's Libtorchvision 0.10.0. Will this difference cause a problem if I'm trying to deploy to both Android and iOS? Do I need to match the version when training, or just not go higher than it?

Thanks!

dipu0 commented 1 year ago

did you find any solution??

jimjam-slam commented 1 year ago

@dipu0 I'm hoping to get to the bottom of it, but this is for a side project, so I only get to work on it from time to tine!

Srini138 commented 8 months ago

Hi @jimjam-slam did u got solution , since i trained custom model on latest torch version after exporting and running in flutters its getting error, did you got answer for the question?

jimjam-slam commented 8 months ago

Hey @Srini138, unfortunately I put this project down for a while and decided to focus on native iOS development—it just felt like I was trying to juggle too many moving pieces! But when I've experimented with other Flutter PyTorch interfaces, I've generally found you need to line up the PyTorch version you use for training and the one you use for interference exactly.