One Model Per View vs One Model, Similar Views (Headfixed Setups) #808
jmdelahanty
started this conversation in
General
Replies: 1 comment 1 reply
-
Hi @jmdelahanty, To use the same model in both views, you would need to merge the projects together as a new model is trained from scratch each iteration using the user-labeled frames in the project. The views are similar enough that its seems plausible to use labels from one view/project to predict on the new view/merged-project. Definitely worth a shot if you already have ample labeled frames from the old view! Let us know how it goes! Thanks, |
Beta Was this translation helpful? Give feedback.
1 reply
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Hey everyone!
I was wondering about whether or not it's possible to use SLEAP for running inference on multiple views if the views are quite similar/the same points are all visible. This is for an application in a headfixed setting for mice.
We're in the process of adapting the work from Nadine Gogolla's lab that can be found here.
In future recordings we're going to have things oriented either the same way as in their paper for one view or close to it. Here's a schematic from their supplemental figure:
I know that we could train a model for each camera view, but since the data is so similar across views I figured it would be worth trying out whether or not a model could successfully train/do inference for multiple views so long as all the points are visible between them. Intuitively it seems nice to have one model with very similar views because:
I'm guessing the advice will be, "Test it out and see what happens!" But just in case this is a bad idea I wanted to reach out and ask.
Beta Was this translation helpful? Give feedback.
All reactions