Deployment Platform: SeeMe.ai

Small update [SOLVED]:

Added support for Ranger and EfficientNet-Pytorch last night to fix the problem. Inference on fp16 needs some investigating…

1 Like

Sharing the question here, @zerotosingularity I was a little confused during my first try as well-I’ve clarified the approach in the thread but could you please add some details/clarifications in the starter notebook?

Thanks!

1 Like

Thank you, @init_27!

I have updated the forum posts as well as the documentation on Github, hope it’s clearer now.

Our iOS app has been approved for beta testing!

If you have completed the quick guide, feel free to get in touch at ios@seeme.ai so you can also start using your model on mobile.

1 Like

I am getting the same output. Does it not work with fastai2? I am using it for a multilabel classification problem.

1 Like

@zerotosingularity can you help @dipam7 ?

1 Like

@ttsantos already fixed it yesterday! :slight_smile:

I got a direct message but did not see the message here for some reason.

The issue was deploying a ‘multilabel image classifier’ which is not yet supported, but I fixed a workaround, so he can use the predictions (but confidence is not working atm).

Thanks for the heads up!!

1 Like

Just released a new version of the iOS for testing.

From the releases notes:

This release contains two major changes:

  1. Public model feed

These models will be available to anyone who installs the app. Your models will show up in this feed if you mark them as ‘public’.

Our first model recognizes all the UN nation flags, and there are some more in the pipeline. That being said, we can’t wait to show your models in here too!

  1. Follow up Actions

Once a prediction is made, you can use the outcome to search Google or Wikipedia. Later on, you will be able to add custom follow up actions, like showing your product page for the recognized product.

Enjoy!

Let me know if you want to try or have created a model you want to share publicly.

It’s been a while since I’ve updated this thread, but our iOS and Android apps are now live (and free) to be used:

iOS:

Android:

For iOS, we automatically convert your model to CoreML, so you can install it locally. Have a look at the updated quick guides to get you going: https://github.com/SeeMe-ai/fastai-quick-guides

1 Like

Just added support for the v2.0.13 release:

Fastest way to deploy your model, including mobile :slight_smile:

1 Like

Quick update:

Hope to show a more complete demo soon…

We added support for the latest v2.1.9, as we always do. However, to upload your v2.1.9 model, you need to manually set the version, because of a small bug (already fixed for v2.1.10):

print(fastai.__version__)
> 2.1.8 

So set the fastai version manually like below:

# SeeMe support different types of applications, with different frameworks and versions.

# Get the application_id for your framework (version).

# Do this
application_id = client.get_application_id(
    base_framework="pytorch",
    framework="fastai",
    base_framework_version=str(torch.__version__),
    framework_version="2.1.9",
    application="image_classification"
)

# Instead of the default (should be fixed in v2.1.10)

application_id = client.get_application_id(
    base_framework="pytorch",
    framework="fastai",
    base_framework_version=str(torch.__version__),
    framework_version=str(fastai.__version__),
    application="image_classification"
)

Have you considered https://pytorch.org/mobile/android/?

It’s on the list for 2021:grinning:

Meaning:

  • automatically convert on upload
  • try/verify/auotmatate optimizations
  • downloadable via SDK/API
  • benchmark vs CoreML/TFLite for size/speed/accuracy
  • include in iOS/Android app

A good day to return from a small home renovation holiday. :slight_smile:

Added support for v2.2.0, v2.2.1, v2.2.2.

Obviously, I’d recommend v2.2.2, but if you really need them, the others are available.

Added support for the newly released v2.3.1, with both PyTorch 1.7.1 and 1.8.1 support.

Personal reminder: I need to update the quick guide on SeeMe.ai Deployment. | Practical Deep Learning for Coders to include:

  • the automatic conversion to TFLite models,
  • web uploads
  • model version management
  • Data flywheel to sync and incorporate model predictions
  • dataset mgmt

HI @zerotosingularity I have been trying to deploy my model. While getting the application_id for framework, I am getting the NotImplementedError. Kindly help

Hi @adityaojha075 ,

thanks for trying our beta and reaching out. I’ve sent you an email to help you get your model deployed!

Help is on the way. :slight_smile:

Thanks @zerotosingularity ,

I have replied you on the mail. Looking forward to.

We’ve added documentation for the Python SDK: Python SDK | SeeMe.ai Docs

It gives a better insight into what we have been working on:

Models:

  • Create
  • Upload
  • Try
  • Deploy
  • Download
  • Version

Datasets:

  • Create
  • Version
  • Manage (labels, data splits)
  • Upload/Download items (images, text)
  • Import/Export dataset versions

Hope it helps, and looking forward to learn from your feedback!