3D flex training with 3DVA components as input for latent dimensions

3D Flexible Refinement
1

Hi all.

Just trying some 3Dflex and from the tutorial I could not work out how to run Flex training with 3DVA-derived coordinates as an input varible.

What is the workflow/order of operations required? I’ve already done my 3DVA and 3DVA display (in the 3 different output types).

Do I take my particle stack and volume from the 3DVA job and prepare these with Flex data prep, and then use these prepared particles as input? Will they keep the 3DVA-derived coordinate info if i do this?

Thanks
Matt

2

Hi @MattF,

Welcome to the forum! To initialize flex training from 3DVA latent coordinates, you can do the following:

  • (If not done already) Run a 3D Flex Data Prep job using the particles & volume from the source refinement job
  • (If not done already) Run a 3D Variability job using the particles & volume from the same source refinement job
  • Create a 3D Flex Training job, then:
    • Connect the Prepared particles from 3D Flex Data Prep to 3D Flex Training.
    • Connect the Flex Mesh from the Flex Mesh Prep job to 3D Flex Training.
    • While still building the 3D Flex Training job, navigate to the Outputs tab of the 3D Variability job. Drag/drop the specific components_mode_x groups under the “All particles” output, into the components_mode_x input of the particles group in the 3D Flex Training job.
    • Set the “Initialize latents from input” parameter to True
    • Set the “Initialize latents input indices” to a comma separated list of component numbers (i.e. the set of x values in components_mode_x that were connected to the 3D Flex Training job)

If done correctly, you should see something in the event log looking like:

  Input particles already have associated latent coordinates.
  Parameter "Initialize latents from input" was set. Initializing latent coordinates from input.
  "Initialize latents input indices" was specified. Using input components [0, 1, 2]

Here, I used the first 3 components (0, 1, and 2) from 3DVA.

Best,
Michael

1 Like
3

Ah I was missing the drag+drop part. Got a few things to try now.

Thanks!