MobileSAM
MobileSAM copied to clipboard
Getting mobileSAM into a format that runs on mobile (tflite, torchscript)
Hi mobileSAM creators!
Thanks for the project, it's really awesome.
Maybe this is my stupidity, but I thought I should highlight this: I've essentially tried and failed over a couple of days to get mobileSAM into either torchscript or tflite. Possible that direct conversion to e.g. CoreML would be painless, haven't tried that just yet.
A lot of this isn't traceable to you: converting SAM for use on mobile appears to be non-trivial.
But: one suggestion for the future, if you can spare the time: offer links/walkthrough for getting this project in shape for use in mobile apps.
Or, if that exists, I'd love to follow it..!
Thanks again,
Matt
Did you make any progress, maybe we can cooperate.
Hi @cummins-orgs @SkyloveQiu ! If you had a look at EdgeSAM, they implemented a CoreML output. I implemented an integration (closed source for now) of it. The thing is that their license is a lot less permissive.
I'm interested by the CoreML output of MobileSAM as well, but did not investigate yet. Don't hesitate to reach out if you're still interested / already made any progress ;)
Hi both,
After a lot of trouble the project I was working on got cancelled. Maybe coreml was the way to go! The issues I had in swift were either to do with the actual model file or to do with pre-processing fwiw. Former is much more complex/verbose in swift than python
On Thu, Jun 27, 2024 at 8:54 AM, Pierre Mardon @.***(mailto:On Thu, Jun 27, 2024 at 8:54 AM, Pierre Mardon < wrote:
Hi @.(https://github.com/cummins-orgs) @.(https://github.com/SkyloveQiu) ! If you had a look at EdgeSAM, they implemented a CoreML output. I implemented an integration (closed source for now) of it. The thing is that their license is a lot less permissive.
I'm interested by the CoreML output of MobileSAM as well, but did not investigate yet. Don't hesitate to reach out if you're still interested / already made any progress ;)
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>
Ok thanks @cummins-orgs ! I think indeed CoreML is the preferred / simplest path ;)
我已经实现了移动端的部署,并且在我的软件中已经应用一年多了
Hi @kestraltorah ! Could you provide more insight on how you performed your integration please?
使用onnx,编码模型+解码模型
---- Replied Message ---- | From | Pierre @.> | | Date | 08/21/2024 00:11 | | To | ChaoningZhang/MobileSAM @.> | | Cc | kestraltorah @.>, Mention @.> | | Subject | Re: [ChaoningZhang/MobileSAM] Getting mobileSAM into a format that runs on mobile (tflite, torchscript) (Issue #141) |
Hi @kestraltorah ! Could you provide more insight on how you performed your integration please?
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>
Oh right, I didn't even consider it was possible to just use ONNX directly in iOS 🤷♂️ I guess I'll try that, thx !