Downloading Joint track auto instead of the machine version

[TDroogsma]

Hello,

Is it possible to get the old joint track auto version instead of the machine learning version?

Kind regards,
Tamara

[ajensen1234]

We're shut down the source forge containing the original JTA installer.

Is there a feature you're looking for? JTML still involves the exact same edge-detection and optimization routine as JTA, but now you can opt to use a neural network to find specific features in the image.

[TDroogsma]

Thankyou for your quick response.
I am mainly asking, because we would like a to fit a CAD model of a TKA onto fluoroscopy images. My supervisor talked to Scott Banks, and he said that the machine learning version is not accurate enough. Therefore, we wanted to use the old version.

[ajensen1234]

I can't imagine why Scott would say that because the underlying optimization present in JTA (DiRECT + Canny edge detection) is still fully present in JointTrack Machine Learning.

The only difference is, if you desire, you can use a neural network to segment the pixels of the TKA components instead of a canny edge detector. This is an entirely optional step, though, and all the old functionality is still present in its entirety.

I apologize if this seems like I'm pushing back unnecessarily, but I genuinely don't understand why Scott would ever say that, given that he understands everything that I just mentioned.

To achieve JTA functionality, you would simply load your calibration file, images, and 3D models. In “edges” or “dilation” view, you can see what the canny edge detector settings are doing to find the edges of the implant, and can adjust them in the right sidebar of the application. In “settings->optimizer settings”, you can adjust the optimization routine to fit your needs. The optimization that the old JTA was using was called “direct dilation”, and you can adjust the parameters as you would in the old application. Doing this is the same mathematical optimization that was done previously.

[TDroogsma]

Allright, I think we may have been under the impression that we had to use the neural network when using JTML, but if that is optional than there is no problem.

Sorry for the misunderstanding and thankyou for your explanation.