Subscribe
Sign in
Nanonets Airtable Models — Build a no-code AI image organizer with Airtable
Easily build Image Classification Models with Airtable and Nanonets. Add your image urls and labels to Airtable. After connecting with Nanonets, your model will be built and hosted within an hour. Integrating your model is as easy as copying 2 lines of code.
35
Replies
@siddharthgoliya We're currently planning to build Multi label classification and Object detection as well :)
I assume you're using transfer learning on a large pretrained network. What's your base architecture?
@fuzew688 We use a variety of architectures and hyperparameters for every training, auto tuning them based on images uploaded. We use mix of open architectures and some proprietary architectures that work well for specific problems.
Hello fellow hunters,
Thank you for stopping by to have a look at Nanonets' Airtable Integration. I'm one of the co-founders of Nanonets and I would like to give a quick overview of this Airtable Integration. We set out to solve the problem of being able to build no code deep learning models right from your spreadsheets. Through this integration, users can easily build production ready image classification models in a few minutes.To give you a little bit of background, Nanonets is a machine learning API for developers to integrate cutting edge ML into their products.
Let me give you a quick walkthrough of this feature.
1. Assume you have a large number of image urls lying in spreadsheets or a database
2. You just need to clone our base Airtable template and add your image urls here and label around 25-50 images for every unique class that you'd like to classify.
3. Connect your Airtable to Nanonets and that's it! Your model should be trained within an hour.
4. You can constantly improve your model by showing us which predictions we got wrong through Airtable itself.
I would urge you to take a look at the product webpage. We have built the product with a lot of passion and would love to have your feedback on it.
Happy to answer any questions.
Sarthak Jain
@husain_safdari We built this tool with the use of airtable API. Google sheets also provides a similar API so definitely possible. Will plan for an integration. One quick way to do this is there are plugins that sync google sheet and airtable as well.
Why does it take time for the predict step to start working, whats happening behind the curtain
@salim_khan after airtable connects with nanonets, we train a custom model to do predictions. Since this model is custom, it takes time to train it. Hence it takes some time for predict to start working. We keep innovating ways to reduce the training time :)
@salim_khan Hey Salim, we try and train the a custom model based on your images which generally takes a few minutes since it's computationally expensive.
It is a product which can be used by anybody, like if you want to categorise your albums according to the people in it. this gives you a very simple way.
Pros:build a model without any hassle of code.
Cons:Doesnt not yet have integration with other platforms .may be google sheets.
@aseem_saxena Hey Aseem,
Yes, we provide multi label classification on https://app.nanonets.com. Currently the airtable integration does not support this. It's something we wanted to build out depending on the reception.
Hey, this is definitely what we are looking for! Thanks for sharing. Do you have an integration for amazon S3
@pbk Hey Nil, we don't have an S3 integration as yet. However, you can always pull the URLs from s3 and add it to airtable :)
@payalshah21 Hey, currently we have a minimum of 25 examples per category. In our experience, this number is enough to build fairly robust models.
@nitisha0992 Airtable has Attachment field type as well in addition to url. We were thinking of doing that instead of urls when we first started working on it. However upload on airtable was bit slow so thought should stick to urls for now and later if lot of people think it would be useful, will add attachments as well. Any interesting use case you have in mind for uploading your own photos?
Optimistic about future integrations. Product is good.
Pros:Easy to use. Retraining support is great
Cons:Only supports image urls
MoMAR