What is behind the Roboflow 3.0 Instance Segmentation model?

[gjm]

The Default network for on this platform for segmentation seems to be the Roboflow 3.0 Instance Segmentation (Fast) network.
Is there any documentation or paper about what exactly network this is?
I found an overview over different common networks, but it is missing the internal Roboflow 3.0 Instance Segmentation (FAST).

I found also this post but there seems to be no hint about the actual architecture or number of weights…

[leo]

Hi @gjm

As mentioned in the topic you mentioned, we do not publish the network or model we use for Roboflow Train at this time.

@brad

We purposely don’t say because it gives us the flexibility to continue to improve and adapt the models we use behind the scenes over time (or even to change out the model completely or dynamically select the right model based on your dataset or domain).

We understand that there are instances in which you need more information and control over the training process, which is why we have a large GitHub repository of notebooks which you can use to train your models.