Today, we are excited to announce a new annotation feature, the ATOM segmenter.
It's a new one-click annotation feature that can save you incredible amounts of time for instance, semantic or panoptic segmentation - especially when annotating complex shapes or images with a lot of obfuscation. Before we go into the details of exactly how it works and how it can help you, let's look at some results that will (hopefully) get you excited:
Perfect segmentations in one click
How it works
Like other semi-automatic annotation tools we have in Hasty, this is an AI-powered tool that uses a pre-trained model. That means that we've shown an AI model all the public data we could find from which it learned how thousands of different types of annotations should look like. Now, experienced vision AI experts will start having some doubts. After all, there's no way a pre-trained model will always produce excellent results.
They are right, but given the amount of data we have trained on, chances are we have seen something close to your use case. It might not always be one-click annotations, but you will be able to do complex annotations in a couple of clicks, as seen here:
Here, the linesman has a flag causing some confusion for ATOM. However, we get a good annotation in 3 clicks.
We have also taken user feedback into account when developing this new feature. During the beta, one of the things we kept hearing is that users wanted to remove areas that were not relevant. This you can do in ATOM. As an example, let's look at us annotating a streetcar.
Note the red circles - these are "negative clicks" telling ATOM to remove certain areas.
Once again, the experienced annotator or vision AI engineer might be thinking, "Ok, that makes sense, but I don't want to pay for semi-automated tooling". We agree.
ATOM is entirely free to use. If you are new to Hasty, it's important to highlight that when we say free, we mean free. We don't have any user limits, annotation time limits, or data upload limits.
Of course, that raises a question. We are not an NGO. We are not millionaires funding an annotation tool for the good of the world. So what is the hook?
Going beyond semi-automated tooling
Our business model is very straightforward. We think that we can automate a significant percentage of your annotation work. Without going into too much detail (which you can find here), we take the data you annotate and train custom models using that data.
You can then access and see predictions from those models. The more you annotate, the better they get. And that's when we start making money. Essentially, you pay when our custom models provide better annotation automation than what you can do with pre-trained models and semi-automatic tools.
It is also essential to address one of the main fears we hear from new users. The models we train only benefit you, so only you and those you invite can access and draw benefit from the models trained on your data.
Give it a try and let us know what you think
If you do so, feel free to let me know what you think. You can email me at alex(at)hasty.ai.
PS. For clarity, if you are an outsourcing software agency that wants to rent out developers, my email is noreply(at)hasty.ai. I've set that up especially for important inquiries of this nature.